[None][infra] Improve unit test CI coverage

.pre-commit-config.yaml

@@ -1296,7 +1296,6 @@ common-files: &common_files |
         tests/unittest/trt/model/test_nemotron_nas.py |
         tests/unittest/trt/model/test_phi.py |
         tests/unittest/trt/model/test_unet.py |
-        tests/unittest/trt/model_api/profile_utils.py |
         tests/unittest/trt/model_api/test_model_api_multi_gpu.py |
         tests/unittest/trt/model_api/test_model_level_api.py |
         tests/unittest/trt/model_api/test_model_quantization.py |
@@ -2671,7 +2670,6 @@ legacy-files: &legacy_files |
         tests/unittest/trt/model/test_nemotron_nas.py |
         tests/unittest/trt/model/test_phi.py |
         tests/unittest/trt/model/test_unet.py |
-        tests/unittest/trt/model_api/profile_utils.py |
         tests/unittest/trt/model_api/test_model_api_multi_gpu.py |
         tests/unittest/trt/model_api/test_model_level_api.py |
         tests/unittest/trt/model_api/test_model_quantization.py |

jenkins/scripts/cbts/blocks.py

@@ -168,6 +168,11 @@ def _strip_params(s: str) -> str:
     return s.rsplit("[", 1)[0] if "[" in s else s
 
 
+def _normalize_target_path(target: str) -> str:
+    """Normalize directory-style targets without changing the root sentinel."""
+    return target.rstrip("/") or target
+
+
 def _entry_target(entry: str) -> str:
     """Canonical "target" key for indexing/lookup.
 
@@ -176,7 +181,7 @@ def _entry_target(entry: str) -> str:
     `file.py::TestC::test_m[a-b] TIMEOUT (90)` → `file.py::TestC::test_m`
     `file.py -k "kw"`                          → `file.py`
     """
-    return _strip_params(_strip_pytest_options(normalize_test_id(entry)))
+    return _normalize_target_path(_strip_params(_strip_pytest_options(normalize_test_id(entry))))
 
 
 def _target_in_filter_subtree(target: str, filter_prefix: str) -> bool:
@@ -185,6 +190,8 @@ def _target_in_filter_subtree(target: str, filter_prefix: str) -> bool:
     `target` matches when it is `filter_prefix` itself or a descendant of it
     (params / method / file / dir component below) in the pytest tree.
     """
+    target = _normalize_target_path(target)
+    filter_prefix = _normalize_target_path(filter_prefix)
     if target == filter_prefix:
         return True
     return (
@@ -204,6 +211,7 @@ def _path_lookup_anchor(yaml_path: str) -> str:
     cover most blocks. Test files anchor on themselves. A top-level
     helper with no enclosing dir returns "" — caller treats as no-match.
     """
+    yaml_path = _normalize_target_path(yaml_path)
     base = yaml_path.rsplit("/", 1)[-1]
     if not base.startswith("test_"):
         return yaml_path.rsplit("/", 1)[0] if "/" in yaml_path else ""
@@ -256,7 +264,7 @@ def _load_one(self, yml_path: Path) -> None:
             for test in tests:
                 seen: set[str] = set()
                 norm = normalize_test_id(test)
-                canonical = _strip_params(_strip_pytest_options(norm))
+                canonical = _normalize_target_path(_strip_params(_strip_pytest_options(norm)))
                 for key in (test, norm, _strip_pytest_options(norm), canonical):
                     if key and key not in seen:
                         seen.add(key)

legacy-files.txt

@@ -1288,7 +1288,6 @@ tests/unittest/trt/model/test_mistral.py
 tests/unittest/trt/model/test_nemotron_nas.py
 tests/unittest/trt/model/test_phi.py
 tests/unittest/trt/model/test_unet.py
-tests/unittest/trt/model_api/profile_utils.py
 tests/unittest/trt/model_api/test_model_api_multi_gpu.py
 tests/unittest/trt/model_api/test_model_level_api.py
 tests/unittest/trt/model_api/test_model_quantization.py

pyproject.toml

@@ -1345,7 +1345,6 @@ exclude = [
     "tests/unittest/trt/model/test_nemotron_nas.py",
     "tests/unittest/trt/model/test_phi.py",
     "tests/unittest/trt/model/test_unet.py",
-    "tests/unittest/trt/model_api/profile_utils.py",
     "tests/unittest/trt/model_api/test_model_api_multi_gpu.py",
     "tests/unittest/trt/model_api/test_model_level_api.py",
     "tests/unittest/trt/model_api/test_model_quantization.py",

ruff-legacy-baseline.json

@@ -2298,9 +2298,6 @@
     "F403": 1,
     "F405": 13
   },
-  "tests/unittest/trt/model_api/profile_utils.py": {
-    "D415": 1
-  },
   "tests/unittest/trt/model_api/test_model_api_multi_gpu.py": {
     "D200": 1,
     "D300": 1,

ruff-legacy.toml

@@ -1305,7 +1305,6 @@ include = [
     "tests/unittest/trt/model/test_nemotron_nas.py",
     "tests/unittest/trt/model/test_phi.py",
     "tests/unittest/trt/model/test_unet.py",
-    "tests/unittest/trt/model_api/profile_utils.py",
     "tests/unittest/trt/model_api/test_model_api_multi_gpu.py",
     "tests/unittest/trt/model_api/test_model_level_api.py",
     "tests/unittest/trt/model_api/test_model_quantization.py",

tests/integration/defs/agg_unit_mem_df.csv

@@ -12,7 +12,7 @@ unittest/trt/attention/test_gpt_attention_IFB.py,NVIDIA A10,17,
 unittest/trt/attention/test_gpt_attention_no_cache.py,NVIDIA A10,23,
 unittest/trt/model/test_mamba.py,NVIDIA A10,12,
 unittest/trt/model/test_llama.py,NVIDIA A10,3,
-unittest/kv_cache_manager_v2_tests/,NVIDIA A10,8,
+unittest/kv_cache_manager_v2_tests,NVIDIA A10,8,
 "unittest/trt/attention/test_gpt_attention.py -k ""partition0""",NVIDIA A10,14,
 "unittest/trt/attention/test_gpt_attention.py -k ""partition1""",NVIDIA A10,10,
 "unittest/trt/attention/test_gpt_attention.py -k ""partition2""",NVIDIA A10,3,
@@ -49,7 +49,7 @@ unittest/trt/quantization/test_weight_only_groupwise_quant_matmul.py,NVIDIA H100
 unittest/trt/attention/test_gpt_attention_IFB.py,NVIDIA H100 80GB HBM3,11,
 unittest/trt/attention/test_gpt_attention_no_cache.py,NVIDIA H100 80GB HBM3,13,
 unittest/trt/model/test_mamba.py,NVIDIA H100 80GB HBM3,10,
-unittest/kv_cache_manager_v2_tests/,NVIDIA H100 80GB HBM3,8,
+unittest/kv_cache_manager_v2_tests,NVIDIA H100 80GB HBM3,8,
 "unittest/trt/attention/test_gpt_attention.py -k ""partition0""",NVIDIA L40S,14,
 "unittest/trt/attention/test_gpt_attention.py -k ""partition1""",NVIDIA L40S,10,
 "unittest/trt/attention/test_gpt_attention.py -k ""partition2""",NVIDIA L40S,6,
@@ -72,7 +72,7 @@ unittest/trt/quantization/test_weight_only_groupwise_quant_matmul.py,NVIDIA H100
 unittest/trt/attention/test_gpt_attention_IFB.py,NVIDIA H100 PCIe,11,
 unittest/trt/attention/test_gpt_attention_no_cache.py,NVIDIA H100 PCIe,13,
 unittest/trt/model/test_mamba.py,NVIDIA H100 PCIe,10,
-unittest/kv_cache_manager_v2_tests/,NVIDIA H100 PCIe,8,
+unittest/kv_cache_manager_v2_tests,NVIDIA H100 PCIe,8,
 llmapi-tp-2gpu,NVIDIA H100 NVL,1,
 unittest/llmapi/test_llm_models_multi_gpu.py,NVIDIA H100 NVL,1,
 unittest/trt/model/test_gptneox.py,NVIDIA H100 NVL,7,
@@ -89,7 +89,7 @@ unittest/trt/quantization/test_weight_only_groupwise_quant_matmul.py,NVIDIA H100
 unittest/trt/attention/test_gpt_attention_IFB.py,NVIDIA H100 NVL,11,
 unittest/trt/attention/test_gpt_attention_no_cache.py,NVIDIA H100 NVL,13,
 unittest/trt/model/test_mamba.py,NVIDIA H100 NVL,10,
-unittest/kv_cache_manager_v2_tests/,NVIDIA H100 NVL,8,
+unittest/kv_cache_manager_v2_tests,NVIDIA H100 NVL,8,
 llmapi-tp-2gpu,NVIDIA H100,1,
 unittest/llmapi/test_llm_models_multi_gpu.py,NVIDIA H100,1,
 unittest/trt/model/test_gptneox.py,NVIDIA H100,7,
@@ -106,7 +106,7 @@ unittest/trt/quantization/test_weight_only_groupwise_quant_matmul.py,NVIDIA H100
 unittest/trt/attention/test_gpt_attention_IFB.py,NVIDIA H100,11,
 unittest/trt/attention/test_gpt_attention_no_cache.py,NVIDIA H100,13,
 unittest/trt/model/test_mamba.py,NVIDIA H100,10,
-unittest/kv_cache_manager_v2_tests/,NVIDIA H100,8,
+unittest/kv_cache_manager_v2_tests,NVIDIA H100,8,
 "unittest/trt/attention/test_gpt_attention.py -k ""partition0""",NVIDIA L40,14,
 "unittest/trt/attention/test_gpt_attention.py -k ""partition1""",NVIDIA L40,10,
 "unittest/trt/attention/test_gpt_attention.py -k ""partition2""",NVIDIA L40,6,
@@ -126,7 +126,7 @@ unittest/_torch/attention,NVIDIA B200,4,
 unittest/_torch/misc,NVIDIA B200,4,
 unittest/_torch/speculative/test_eagle3.py,NVIDIA B200,4,
 unittest/_torch/thop/parallel,NVIDIA B200,16,
-unittest/kv_cache_manager_v2_tests/,NVIDIA B200,8,
+unittest/kv_cache_manager_v2_tests,NVIDIA B200,8,
 unittest/auto_deploy/singlegpu/compile,NVIDIA H100,4,
 unittest/auto_deploy/singlegpu/custom_ops,NVIDIA H100,4,
 unittest/auto_deploy/singlegpu/models,NVIDIA H100,4,

tests/integration/test_lists/test-db/l0_a10.yml

@@ -35,16 +35,15 @@ l0_a10:
   - unittest/_torch/executor/test_kv_cache_budget_split.py
   - unittest/_torch/executor/test_kv_pool_rebalance.py
   - unittest/_torch/executor/test_disagg_index_mapper_early_release.py
-  - unittest/_torch/pyexecutor/test_kv_cache_compression_manager.py
-  - unittest/_torch/pyexecutor/test_error_classification.py
+  - unittest/_torch/executor/test_kv_cache_compression_manager.py
+  - unittest/_torch/executor/test_error_classification.py
   - unittest/_torch/modules/dwdp/test_dwdp_fixup_moe_backends.py
   - unittest/_torch/modules/dwdp/test_dwdp_manager.py
   - unittest/_torch/modules/dwdp/test_dwdp_mapping.py
   - unittest/_torch/modules/dwdp/test_dwdp_peer_ranges.py
   # NOTE: this is a CPU-only test, but we do not have a dedicated job for this (and therefore no
   # test list either).
-  - unittest/_torch/models/checkpoints/hf/test_weight_loader.py
-  - unittest/_torch/models/checkpoints/hf/test_checkpoint_loader.py
+  - unittest/_torch/models/checkpoints
   - unittest/_torch/weight_sharing
   - unittest/inputs/test_chat_template_dispatch.py
   - unittest/inputs/test_content_format.py
@@ -70,6 +69,7 @@ l0_a10:
   - unittest/disaggregated/test_peer.py
   - unittest/disaggregated/region/test_block.py
   - unittest/disaggregated/test_mamba_transfer.py
+  - unittest/tools
   - unittest/usage/test_collectors.py
   - unittest/usage/test_config.py
   - unittest/usage/test_opt_out.py
@@ -204,7 +204,7 @@ l0_a10:
       backend: tensorrt
   tests:
   # ------------- TRT tests ---------------
-  - unittest/kv_cache_manager_v2_tests/ # 4 min
+  - unittest/kv_cache_manager_v2_tests # 4 min
   - unittest/dynamo
   - unittest/api_stability
   - unittest/bindings
@@ -247,6 +247,7 @@ l0_a10:
   - test_e2e.py::test_trtllm_bench_sanity[--non-streaming-FP16-meta-llama/Llama-3.1-8B-llama-3.1-model/Meta-Llama-3.1-8B]
   - test_e2e.py::test_trtllm_bench_latency_sanity[FP16-meta-llama/Llama-3.1-8B-llama-3.1-model/Meta-Llama-3.1-8B]
   - unittest/trt/quantization
+  - unittest/trt/python_plugin
   - unittest/trt/functional # 37 mins
   - llmapi/test_llm_examples.py::test_llmapi_quickstart_atexit
   - unittest/test_model_runner_cpp.py

tests/integration/test_lists/test-db/l0_a100.yml

@@ -77,6 +77,7 @@ l0_a100:
   - unittest/trt/model_api/test_model_api_multi_gpu.py
   - unittest/trt/model/test_gpt_e2e.py
   - unittest/trt/model/eagle
+  - unittest/trt/model/redrafter
   - unittest/llmapi/test_llm_models.py -m "part0"
   - examples/test_whisper.py::test_llm_whisper_general[large-v3-disable_gemm_plugin-enable_attention_plugin-disable_weight_only-float16-nb:1-use_cpp_runtime]
   - examples/test_whisper.py::test_llm_whisper_general[large-v3-disable_gemm_plugin-enable_attention_plugin-disable_weight_only-float16-nb:1-use_python_runtime]

tests/integration/test_lists/test-db/l0_a30.yml

@@ -204,6 +204,7 @@ l0_a30:
   tests:
   - unittest/auto_deploy/singlegpu/compile
   - unittest/auto_deploy/singlegpu/custom_ops
+  - unittest/auto_deploy/singlegpu/mlir
   - unittest/auto_deploy/singlegpu/models
   - unittest/auto_deploy/singlegpu/shim
   - unittest/auto_deploy/singlegpu/smoke

tests/integration/test_lists/test-db/l0_b200.yml

@@ -173,7 +173,7 @@ l0_b200:
   - unittest/tools/test_layer_wise_benchmarks.py::test_nemotron_gen_dep[1]
   - unittest/tools/test_layer_wise_benchmarks.py::test_qwen3_next_gen_tep[1]
   - unittest/tools/test_layer_wise_benchmarks.py::test_performance_alignment[1]
-  - unittest/kv_cache_manager_v2_tests/
+  - unittest/kv_cache_manager_v2_tests
   # ------------- KV Cache V2 Scheduler IT ---------------
   - kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_v2_vs_v1_basic
   - kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_token_budget_limited

tests/integration/test_lists/test-db/l0_dgx_h100.yml

@@ -18,6 +18,7 @@ l0_dgx_h100:
   - unittest/llmapi/test_llm_multi_gpu_pytorch.py -m "gpu2"
   - unittest/llmapi/test_additional_model_outputs.py -m "gpu2"
   - unittest/_torch/multi_gpu -m "not post_merge" TIMEOUT (90)
+  - unittest/_torch/distributed
   - unittest/_torch/modeling/test_modeling_pixtral.py::test_tensor_parallelism
   # ------------- Disaggregated serving tests ---------------
   - accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_eagle3[eagle3_one_model=True-overlap_scheduler=True]
@@ -376,6 +377,7 @@ l0_dgx_h100:
       auto_trigger: others
       orchestrator: mpi
   tests:
+  - unittest/auto_deploy/multigpu/compile
   - unittest/auto_deploy/multigpu/custom_ops
   - unittest/auto_deploy/multigpu/smoke
   - unittest/auto_deploy/multigpu/transformations

tests/integration/test_lists/test-db/l0_h100.yml

@@ -16,10 +16,12 @@ l0_h100:
   tests:
   # ------------- PyTorch tests ---------------
   - unittest/_torch/attention
+  - unittest/_torch/auto_deploy
   - unittest/_torch/compilation
   - unittest/_torch/debugger
   - unittest/_torch/executor
   - unittest/_torch/lora
+  - unittest/_torch/memory
   - unittest/_torch/misc
   # ------------- modules (non-MoE) ---------------
   - unittest/_torch/modules/test_mla_helix.py
@@ -94,6 +96,7 @@ l0_h100:
   - unittest/others/test_kv_cache_transceiver.py::test_kv_cache_transceiver_single_process[PYTHON-mha-ctx_fp16_gen_fp16]
   - unittest/others/test_kv_cache_transceiver.py::test_kv_cache_transceiver_single_process[PYTHON-mla-ctx_fp16_gen_fp16]
   - unittest/llmapi/test_llm_telemetry.py
+  - unittest/scaffolding
   - unittest/usage/test_collectors.py
   - unittest/usage/test_config.py
   - unittest/usage/test_opt_out.py
@@ -312,6 +315,7 @@ l0_h100:
   - test_e2e.py::test_trtllm_bench_iteration_log[TRT-streaming-meta-llama/Llama-3.1-8B-llama-3.1-model/Meta-Llama-3.1-8B]
   - examples/test_qwen.py::test_llm_hf_qwen_multi_lora_1gpu[qwen2.5_1.5b_instruct]
   - unittest/trt/model/eagle # 1 mins on H100
+  - unittest/trt/model/redrafter
   - unittest/llmapi/test_llm_quant.py # 5.5 mins on H100
   - examples/visual_gen/test_visual_gen.py::test_visual_gen_quickstart
   - examples/visual_gen/test_visual_gen.py::test_visual_gen_api_walkthrough
@@ -495,7 +499,7 @@ l0_h100:
   - unittest/trt/model/test_gpt_e2e.py # 3 mins / 6 mins on H100
   - unittest/trt/attention/test_gpt_attention_no_cache.py
   - examples/test_gpt.py::test_gpt_oss_20b_lora_torch[gpt-oss-20b-lora-adapter_NIM_r8-gpt-oss-20b]
-  - unittest/kv_cache_manager_v2_tests/ # 4 min
+  - unittest/kv_cache_manager_v2_tests # 4 min
   # ------------- KV Cache Iteration Stats ---------------
   - unittest/executor/test_stats_serializer.py
   - unittest/metrics/test_collector.py
@@ -538,6 +542,7 @@ l0_h100:
   tests:
   - unittest/auto_deploy/singlegpu/compile
   - unittest/auto_deploy/singlegpu/custom_ops
+  - unittest/auto_deploy/singlegpu/mlir
   - unittest/auto_deploy/singlegpu/models
   - unittest/auto_deploy/singlegpu/shim
   - unittest/auto_deploy/singlegpu/smoke

tests/integration/test_lists/test-db/l0_sanity_check.yml

@@ -37,8 +37,8 @@ l0_sanity_check:
     - unittest/others/test_kv_cache_transceiver.py::test_async_transfer_keeps_llm_request_alive
     - unittest/others/test_kv_cache_transceiver.py::test_kv_transfer_timeout_warns_once_per_request
     - unittest/others/test_kv_cache_transceiver.py::test_kv_transfer_timeout_silent_when_unset
-    - unittest/_torch/pyexecutor/test_model_loader_mx.py
-    - unittest/_torch/pyexecutor/test_hang_detector_kill.py
+    - unittest/_torch/executor/test_model_loader_mx.py
+    - unittest/_torch/executor/test_hang_detector_kill.py
 - condition:
     ranges:
       system_gpu_count:

tests/unittest/_torch/attention_backend/test_attention_op_sync.py → tests/unittest/_torch/attention/test_attention_op_sync.py

@@ -42,6 +42,8 @@
 import typing
 from dataclasses import fields
 
+import pytest
+
 from tensorrt_llm._torch.attention_backend.fmha.fallback import (
     _THOP_EXCLUDED_FIELDS,
     _THOP_LITERALS,
@@ -78,6 +80,7 @@
 # The C++ attention() declaration is the single source of truth for kwarg
 # names, ordering, and types.
 _HEADER = pathlib.Path(__file__).resolve().parents[4] / ("cpp/tensorrt_llm/thop/attentionOp.h")
+_THOP_SYNC_NVBUG = pytest.mark.skip(reason="https://nvbugs/6336801")
 
 
 # ---- C++ declaration parser -------------------------------------------------
@@ -456,6 +459,7 @@ def test_each_source_attr_kwarg_resolves_uniquely():
                 )
 
 
+@_THOP_SYNC_NVBUG
 def test_attr_kwarg_names_match_source_leaf_attrs_except_allowlisted_aliases():
     """Most ``thop.attention`` kwargs should bind to a source attribute with
     the same name. Existing aliases must stay explicit so new semantic
@@ -556,6 +560,7 @@ def _verify_consumed(cls, chains: set[tuple[str, ...]], excluded=frozenset()):
             )
 
 
+@_THOP_SYNC_NVBUG
 def test_every_forward_args_field_is_consumed():
     """Recursively check that every dataclass field reachable from
     ``AttentionForwardArgs`` (including nested sub-bags such as