[TRTLLM-14839][chore] Add shim files for root PEFT module relocation

.github/CODEOWNERS

@@ -462,6 +462,16 @@
 /tensorrt_llm/_torch/pyexecutor/py_executor.py @NVIDIA/trt-llm-runtime-devs @NVIDIA/trt-llm-disagg-devs
 /tensorrt_llm/_torch/pyexecutor/model_engine.py @NVIDIA/trt-llm-runtime-devs @NVIDIA/trt-llm-disagg-devs
 
+# ===== COMPATIBILITY FORWARDING MODULES =====
+# Definition-free re-export modules left at a retired import path so external
+# callers keep working through the migration window (Epic TRTLLM-14558).  Each
+# one keeps the owner its pre-move path had -- without these lines the paths
+# fall through to the repository default, which would be an ownership change
+# nobody asked for.  Removed together with the modules themselves.
+/tensorrt_llm/lora_helper.py @NVIDIA/trt-llm-runtime-devs
+/tensorrt_llm/lora_manager.py @NVIDIA/trt-llm-runtime-devs
+/tensorrt_llm/prompt_adapter_manager.py @NVIDIA/trt-llm-runtime-devs
+
 # ===== GOVERNANCE / LEGAL GATES =====
 /.github/CODEOWNERS @NVIDIA/trt-llm-infra-devs @NVIDIA/trt-llm-oss-compliance
 /.github/tava_architecture_diagram.md @NVIDIA/trt-llm-TAVA-design-change

tensorrt_llm/llmapi/serialization.py

@@ -121,6 +121,16 @@
     ],
     "torch._utils": ["_rebuild_tensor_v2"],
     "torch.storage": ["_load_from_bytes"],
+    # Pre-move module paths, so older artifacts still load.  find_class()
+    # matches the key exactly and rejects before the forwarding module is
+    # imported, so the shims alone do not cover this path.  Removed together
+    # with the forwarding modules.
+    #
+    # lora_helper is the module a real pre-move LoraConfig pickle names -- it
+    # was absent here, so those artifacts have never loaded.  lora_manager is
+    # the key that was present; it is kept as it was.
+    "tensorrt_llm.lora_helper": ["LoraConfig"],
+    "tensorrt_llm.lora_manager": ["LoraConfig"],
 }
 
 

tensorrt_llm/lora_helper.py

@@ -0,0 +1,43 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Compatibility shim for ``tensorrt_llm.lora_helper``.
+
+Will be removed once all usages are migrated to
+``tensorrt_llm._torch.peft.lora.config``.
+
+DO NOT ADD ANYTHING TO THIS FILE.
+"""
+
+import warnings
+
+from tensorrt_llm._torch.peft.lora.config import (  # noqa: F401
+    LoraConfig,
+    get_default_trtllm_modules_to_hf_modules,
+    get_missing_qkv_modules_from_lora_modules,
+)
+
+warnings.warn(
+    "tensorrt_llm.lora_helper has moved to "
+    "tensorrt_llm._torch.peft.lora.config and will be removed in a future "
+    "release.",
+    FutureWarning,
+    stacklevel=2,
+)
+
+__all__ = [
+    "LoraConfig",
+    "get_default_trtllm_modules_to_hf_modules",
+    "get_missing_qkv_modules_from_lora_modules",
+]

tensorrt_llm/lora_manager.py

@@ -0,0 +1,54 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Compatibility shim for ``tensorrt_llm.lora_manager``.
+
+Will be removed once all usages are migrated to
+``tensorrt_llm._torch.peft.lora.config`` and
+``tensorrt_llm._torch.peft.lora.loaders`` and
+``tensorrt_llm._torch.peft.lora.manager``.
+
+DO NOT ADD ANYTHING TO THIS FILE.
+"""
+
+import warnings
+
+from tensorrt_llm._torch.peft.lora.config import LoraConfig  # noqa: F401
+from tensorrt_llm._torch.peft.lora.loaders import HfLoraLoader, NemoLoraLoader  # noqa: F401
+from tensorrt_llm._torch.peft.lora.manager import (  # noqa: F401
+    LoraManager,
+    LoraModelConfig,
+    load_torch_lora,
+    load_torch_nemo_lora,
+)
+
+warnings.warn(
+    "tensorrt_llm.lora_manager has moved to "
+    "tensorrt_llm._torch.peft.lora.config and "
+    "tensorrt_llm._torch.peft.lora.loaders and "
+    "tensorrt_llm._torch.peft.lora.manager and will be removed in a "
+    "future release.",
+    FutureWarning,
+    stacklevel=2,
+)
+
+__all__ = [
+    "HfLoraLoader",
+    "LoraConfig",
+    "LoraManager",
+    "LoraModelConfig",
+    "NemoLoraLoader",
+    "load_torch_lora",
+    "load_torch_nemo_lora",
+]

tensorrt_llm/prompt_adapter_manager.py

@@ -0,0 +1,37 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Compatibility shim for ``tensorrt_llm.prompt_adapter_manager``.
+
+Will be removed once all usages are migrated to
+``tensorrt_llm._torch.peft.prompt_adapter``.
+
+DO NOT ADD ANYTHING TO THIS FILE.
+"""
+
+import warnings
+
+from tensorrt_llm._torch.peft.prompt_adapter import PromptAdapterManager  # noqa: F401
+
+warnings.warn(
+    "tensorrt_llm.prompt_adapter_manager has moved to "
+    "tensorrt_llm._torch.peft.prompt_adapter and will be removed in a "
+    "future release.",
+    FutureWarning,
+    stacklevel=2,
+)
+
+__all__ = [
+    "PromptAdapterManager",
+]

tests/unittest/_torch/peft/test_shim_peft.py

@@ -0,0 +1,99 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""The compatibility modules kept at the pre-move PEFT import paths.
+
+Each one must forward every name in its own ``__all__`` to the module that
+defines it, as the same object.  The names are read from the module rather than
+restated here, so changing the export set changes what is verified.
+
+``lora_manager`` is the interesting case: it is the only one whose names do not
+all live in a single module.  It was one file before the move and is three now,
+so the test resolves each name to wherever it is actually defined instead of
+assuming a single destination -- which is also what stops the shim from
+quietly pointing a name at a module that merely re-exports it.
+"""
+
+import importlib
+import sys
+import warnings
+
+import pytest
+
+pytestmark = pytest.mark.cpu_only
+
+FORWARDING_PATHS = (
+    "tensorrt_llm.lora_helper",
+    "tensorrt_llm.lora_manager",
+    "tensorrt_llm.prompt_adapter_manager",
+)
+
+CANONICAL_PREFIX = "tensorrt_llm._torch.peft."
+
+
+@pytest.mark.parametrize("forwarding_path", FORWARDING_PATHS)
+def test_published_names_come_from_where_they_are_defined(forwarding_path: str) -> None:
+    """Identity against the defining module, not against the shim's target.
+
+    A re-implementation would compare equal and still break ``isinstance`` and
+    unpickling for callers that kept the old import path.  Resolving through
+    ``__module__`` additionally catches a shim that forwards a name to a module
+    that only re-exports it: that spelling resolves, so nothing fails until the
+    incidental re-export goes away.
+    """
+    forwarding = importlib.import_module(forwarding_path)
+
+    published = getattr(forwarding, "__all__", None)
+    assert published, f"{forwarding_path} publishes nothing, so it forwards nothing"
+
+    for name in published:
+        obj = getattr(forwarding, name)
+        home = getattr(obj, "__module__", None)
+        assert home is not None, f"{name} carries no __module__ to check"
+        assert home.startswith(CANONICAL_PREFIX), f"{name}.__module__ is {home!r}"
+        assert getattr(importlib.import_module(home), name) is obj, name
+
+
+@pytest.mark.parametrize("forwarding_path", FORWARDING_PATHS)
+def test_import_warns_once_and_names_the_new_path(forwarding_path: str) -> None:
+    """Importing the old path has to say so, and say where to go instead."""
+    # A module body runs once per interpreter, so drop it before re-importing.
+    sys.modules.pop(forwarding_path, None)
+    with warnings.catch_warnings(record=True) as caught:
+        warnings.simplefilter("always")
+        forwarding = importlib.import_module(forwarding_path)
+
+    about_this_module = [w for w in caught if forwarding_path in str(w.message)]
+    assert len(about_this_module) == 1, [str(w.message) for w in caught]
+
+    message = str(about_this_module[0].message)
+    for name in forwarding.__all__:
+        home = getattr(getattr(forwarding, name), "__module__", "")
+        assert home in message, f"the warning does not mention {home}"
+
+
+def test_the_restricted_unpickle_allowlist_still_accepts_the_old_module_strings() -> None:
+    """``find_class`` matches the key exactly, before any import happens.
+
+    So the forwarding modules alone do not cover deserialization: a pickle
+    written before the move names the old module, and a canonical-only table
+    rejects it without ever reaching the shim.
+    """
+    from tensorrt_llm.llmapi.serialization import BASE_EXAMPLE_CLASSES
+
+    canonical = "tensorrt_llm._torch.peft.lora.config"
+    assert "LoraConfig" in BASE_EXAMPLE_CLASSES[canonical]
+    # lora_helper is the module a real pre-move LoraConfig pickle names.
+    assert "LoraConfig" in BASE_EXAMPLE_CLASSES["tensorrt_llm.lora_helper"]
+    assert "LoraConfig" in BASE_EXAMPLE_CLASSES["tensorrt_llm.lora_manager"]