feat(llm): Add automatic provider inference for LangChain LLMs (#1460)

Author: Pouyanpi

nemoguardrails/actions/llm/utils.py

@@ -46,6 +46,70 @@ def __init__(self, inner_exception: Any):
         self.inner_exception = inner_exception
 
 
+def _infer_provider_from_module(llm: BaseLanguageModel) -> Optional[str]:
+    """Infer provider name from the LLM's module path.
+
+    This function extracts the provider name from LangChain package naming conventions:
+    - langchain_openai -> openai
+    - langchain_anthropic -> anthropic
+    - langchain_google_genai -> google_genai
+    - langchain_nvidia_ai_endpoints -> nvidia_ai_endpoints
+    - langchain_community.chat_models.ollama -> ollama
+
+    For patched/wrapped classes, checks base classes as well.
+
+    Args:
+        llm: The LLM instance
+
+    Returns:
+        The inferred provider name, or None if it cannot be determined
+    """
+    module = type(llm).__module__
+
+    if module.startswith("langchain_"):
+        package = module.split(".")[0]
+        provider = package.replace("langchain_", "")
+
+        if provider == "community":
+            parts = module.split(".")
+            if len(parts) >= 3:
+                provider = parts[-1]
+                return provider
+        else:
+            return provider
+
+    for base_class in type(llm).__mro__[1:]:
+        base_module = base_class.__module__
+        if base_module.startswith("langchain_"):
+            package = base_module.split(".")[0]
+            provider = package.replace("langchain_", "")
+
+            if provider == "community":
+                parts = base_module.split(".")
+                if len(parts) >= 3:
+                    provider = parts[-1]
+                    return provider
+            else:
+                return provider
+
+    return None
+
+
+def get_llm_provider(llm: BaseLanguageModel) -> Optional[str]:
+    """Get the provider name for an LLM instance by inferring from module path.
+
+    This function extracts the provider name from LangChain package naming conventions.
+    See _infer_provider_from_module for details on the inference logic.
+
+    Args:
+        llm: The LLM instance
+
+    Returns:
+        The provider name if it can be inferred, None otherwise
+    """
+    return _infer_provider_from_module(llm)
+
+
 def _infer_model_name(llm: BaseLanguageModel):
     """Helper to infer the model name based from an LLM instance.
 
@@ -126,7 +190,7 @@ def _setup_llm_call_info(
         llm_call_info_var.set(llm_call_info)
 
     llm_call_info.llm_model_name = model_name or _infer_model_name(llm)
-    llm_call_info.llm_provider_name = model_provider
+    llm_call_info.llm_provider_name = model_provider or _infer_provider_from_module(llm)
 
 
 def _prepare_callbacks(

tests/test_actions_llm_utils.py

@@ -0,0 +1,125 @@
+# SPDX-FileCopyrightText: Copyright (c) 2023-2025 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.
+
+from nemoguardrails.actions.llm.utils import _infer_provider_from_module
+
+
+class MockOpenAILLM:
+    __module__ = "langchain_openai.chat_models"
+
+
+class MockAnthropicLLM:
+    __module__ = "langchain_anthropic.chat_models"
+
+
+class MockNVIDIALLM:
+    __module__ = "langchain_nvidia_ai_endpoints.chat_models"
+
+
+class MockCommunityOllama:
+    __module__ = "langchain_community.chat_models.ollama"
+
+
+class MockUnknownLLM:
+    __module__ = "some_custom_package.models"
+
+
+class MockNVIDIAOriginal:
+    __module__ = "langchain_nvidia_ai_endpoints.chat_models"
+
+
+class MockPatchedNVIDIA(MockNVIDIAOriginal):
+    __module__ = "nemoguardrails.llm.providers._langchain_nvidia_ai_endpoints_patch"
+
+
+def test_infer_provider_openai():
+    llm = MockOpenAILLM()
+    provider = _infer_provider_from_module(llm)
+    assert provider == "openai"
+
+
+def test_infer_provider_anthropic():
+    llm = MockAnthropicLLM()
+    provider = _infer_provider_from_module(llm)
+    assert provider == "anthropic"
+
+
+def test_infer_provider_nvidia_ai_endpoints():
+    llm = MockNVIDIALLM()
+    provider = _infer_provider_from_module(llm)
+    assert provider == "nvidia_ai_endpoints"
+
+
+def test_infer_provider_community_ollama():
+    llm = MockCommunityOllama()
+    provider = _infer_provider_from_module(llm)
+    assert provider == "ollama"
+
+
+def test_infer_provider_unknown():
+    llm = MockUnknownLLM()
+    provider = _infer_provider_from_module(llm)
+    assert provider is None
+
+
+def test_infer_provider_from_patched_class():
+    llm = MockPatchedNVIDIA()
+    provider = _infer_provider_from_module(llm)
+    assert provider == "nvidia_ai_endpoints"
+
+
+def test_infer_provider_checks_base_classes():
+    class BaseOpenAI:
+        __module__ = "langchain_openai.chat_models"
+
+    class CustomWrapper(BaseOpenAI):
+        __module__ = "my_custom_wrapper.llms"
+
+    llm = CustomWrapper()
+    provider = _infer_provider_from_module(llm)
+    assert provider == "openai"
+
+
+def test_infer_provider_multiple_inheritance():
+    class BaseNVIDIA:
+        __module__ = "langchain_nvidia_ai_endpoints.chat_models"
+
+    class Mixin:
+        __module__ = "some_mixin.utils"
+
+    class MultipleInheritance(Mixin, BaseNVIDIA):
+        __module__ = "custom_package.models"
+
+    llm = MultipleInheritance()
+    provider = _infer_provider_from_module(llm)
+    assert provider == "nvidia_ai_endpoints"
+
+
+def test_infer_provider_deeply_nested_inheritance():
+    class Original:
+        __module__ = "langchain_anthropic.chat_models"
+
+    class Wrapper1(Original):
+        __module__ = "wrapper1.models"
+
+    class Wrapper2(Wrapper1):
+        __module__ = "wrapper2.models"
+
+    class Wrapper3(Wrapper2):
+        __module__ = "wrapper3.models"
+
+    llm = Wrapper3()
+    provider = _infer_provider_from_module(llm)
+    assert provider == "anthropic"