core,integrations[minor]: Dont error on fields in model_kwargs (#27110)

Given the current erroring behavior, every time we've moved a kwarg from
model_kwargs and made it its own field that was a breaking change.
Updating this behavior to support the old instantiations /
serializations.

Assuming build_extra_kwargs was not something that itself is being used
externally and needs to be kept backwards compatible

libs/community/langchain_community/chat_models/snowflake.py

@@ -17,7 +17,7 @@ from langchain_core.utils import (
     get_pydantic_field_names,
     pre_init,
 )
-from langchain_core.utils.utils import build_extra_kwargs
+from langchain_core.utils.utils import _build_model_kwargs
 from pydantic import Field, SecretStr, model_validator
 
 SUPPORTED_ROLES: List[str] = [
@@ -131,10 +131,7 @@ class ChatSnowflakeCortex(BaseChatModel):
     def build_extra(cls, values: Dict[str, Any]) -> Any:
         """Build extra kwargs from additional params that were passed in."""
         all_required_field_names = get_pydantic_field_names(cls)
-        extra = values.get("model_kwargs", {})
-        values["model_kwargs"] = build_extra_kwargs(
-            extra, values, all_required_field_names
-        )
+        values = _build_model_kwargs(values, all_required_field_names)
         return values
 
     @pre_init

libs/community/langchain_community/llms/anthropic.py

@@ -26,7 +26,7 @@ from langchain_core.utils import (
     get_pydantic_field_names,
     pre_init,
 )
-from langchain_core.utils.utils import build_extra_kwargs, convert_to_secret_str
+from langchain_core.utils.utils import _build_model_kwargs, convert_to_secret_str
 from pydantic import ConfigDict, Field, SecretStr, model_validator
 
 
@@ -69,11 +69,8 @@ class _AnthropicCommon(BaseLanguageModel):
     @model_validator(mode="before")
     @classmethod
     def build_extra(cls, values: Dict) -> Any:
-        extra = values.get("model_kwargs", {})
         all_required_field_names = get_pydantic_field_names(cls)
-        values["model_kwargs"] = build_extra_kwargs(
-            extra, values, all_required_field_names
-        )
+        values = _build_model_kwargs(values, all_required_field_names)
         return values
 
     @pre_init

libs/community/langchain_community/llms/llamacpp.py

@@ -8,7 +8,7 @@ from langchain_core.callbacks import CallbackManagerForLLMRun
 from langchain_core.language_models.llms import LLM
 from langchain_core.outputs import GenerationChunk
 from langchain_core.utils import get_pydantic_field_names, pre_init
-from langchain_core.utils.utils import build_extra_kwargs
+from langchain_core.utils.utils import _build_model_kwargs
 from pydantic import Field, model_validator
 
 logger = logging.getLogger(__name__)
@@ -199,10 +199,7 @@ class LlamaCpp(LLM):
     def build_model_kwargs(cls, values: Dict[str, Any]) -> Any:
         """Build extra kwargs from additional params that were passed in."""
         all_required_field_names = get_pydantic_field_names(cls)
-        extra = values.get("model_kwargs", {})
-        values["model_kwargs"] = build_extra_kwargs(
-            extra, values, all_required_field_names
-        )
+        values = _build_model_kwargs(values, all_required_field_names)
         return values
 
     @property

libs/community/langchain_community/llms/openai.py

@@ -34,7 +34,7 @@ from langchain_core.utils import (
     pre_init,
 )
 from langchain_core.utils.pydantic import get_fields
-from langchain_core.utils.utils import build_extra_kwargs
+from langchain_core.utils.utils import _build_model_kwargs
 from pydantic import ConfigDict, Field, model_validator
 
 from langchain_community.utils.openai import is_openai_v1
@@ -268,10 +268,7 @@ class BaseOpenAI(BaseLLM):
     def build_extra(cls, values: Dict[str, Any]) -> Any:
         """Build extra kwargs from additional params that were passed in."""
         all_required_field_names = get_pydantic_field_names(cls)
-        extra = values.get("model_kwargs", {})
-        values["model_kwargs"] = build_extra_kwargs(
-            extra, values, all_required_field_names
-        )
+        values = _build_model_kwargs(values, all_required_field_names)
         return values
 
     @pre_init

libs/community/tests/unit_tests/chat_models/test_anthropic.py

@@ -33,9 +33,12 @@ def test_anthropic_model_kwargs() -> None:
 
 
 @pytest.mark.requires("anthropic")
-def test_anthropic_invalid_model_kwargs() -> None:
-    with pytest.raises(ValueError):
-        ChatAnthropic(model_kwargs={"max_tokens_to_sample": 5})
+def test_anthropic_fields_in_model_kwargs() -> None:
+    """Test that for backwards compatibility fields can be passed in as model_kwargs."""
+    llm = ChatAnthropic(model_kwargs={"max_tokens_to_sample": 5})
+    assert llm.max_tokens_to_sample == 5
+    llm = ChatAnthropic(model_kwargs={"max_tokens": 5})
+    assert llm.max_tokens_to_sample == 5
 
 
 @pytest.mark.requires("anthropic")

libs/community/tests/unit_tests/llms/test_openai.py

@@ -26,13 +26,12 @@ def test_openai_model_kwargs() -> None:
 
 
 @pytest.mark.requires("openai")
-def test_openai_invalid_model_kwargs() -> None:
-    with pytest.raises(ValueError):
-        OpenAI(model_kwargs={"model_name": "foo"})
-
-    # Test that "model" cannot be specified in kwargs
-    with pytest.raises(ValueError):
-        OpenAI(model_kwargs={"model": "gpt-3.5-turbo-instruct"})
+def test_openai_fields_model_kwargs() -> None:
+    """Test that for backwards compatibility fields can be passed in as model_kwargs."""
+    llm = OpenAI(model_kwargs={"model_name": "foo"}, api_key="foo")
+    assert llm.model_name == "foo"
+    llm = OpenAI(model_kwargs={"model": "foo"}, api_key="foo")
+    assert llm.model_name == "foo"
 
 
 @pytest.mark.requires("openai")

libs/core/langchain_core/utils/__init__.py

@@ -32,6 +32,7 @@ from langchain_core.utils.utils import (
 )
 
 __all__ = [
+    "build_extra_kwargs",
     "StrictFormatter",
     "check_package_version",
     "convert_to_secret_str",
@@ -46,7 +47,6 @@ __all__ = [
     "raise_for_status_with_text",
     "xor_args",
     "try_load_from_hub",
-    "build_extra_kwargs",
     "image",
     "get_from_env",
     "get_from_dict_or_env",

libs/core/langchain_core/utils/utils.py

@@ -210,6 +210,51 @@ def get_pydantic_field_names(pydantic_cls: Any) -> set[str]:
     return all_required_field_names
 
 
+def _build_model_kwargs(
+    values: dict[str, Any],
+    all_required_field_names: set[str],
+) -> dict[str, Any]:
+    """Build "model_kwargs" param from Pydanitc constructor values.
+
+    Args:
+        values: All init args passed in by user.
+        all_required_field_names: All required field names for the pydantic class.
+
+    Returns:
+        Dict[str, Any]: Extra kwargs.
+
+    Raises:
+        ValueError: If a field is specified in both values and extra_kwargs.
+        ValueError: If a field is specified in model_kwargs.
+    """
+    extra_kwargs = values.get("model_kwargs", {})
+    for field_name in list(values):
+        if field_name in extra_kwargs:
+            raise ValueError(f"Found {field_name} supplied twice.")
+        if field_name not in all_required_field_names:
+            warnings.warn(
+                f"""WARNING! {field_name} is not default parameter.
+                {field_name} was transferred to model_kwargs.
+                Please confirm that {field_name} is what you intended.""",
+                stacklevel=7,
+            )
+            extra_kwargs[field_name] = values.pop(field_name)
+
+    invalid_model_kwargs = all_required_field_names.intersection(extra_kwargs.keys())
+    if invalid_model_kwargs:
+        warnings.warn(
+            f"Parameters {invalid_model_kwargs} should be specified explicitly. "
+            f"Instead they were passed in as part of `model_kwargs` parameter.",
+            stacklevel=7,
+        )
+        for k in invalid_model_kwargs:
+            values[k] = extra_kwargs.pop(k)
+
+    values["model_kwargs"] = extra_kwargs
+    return values
+
+
+# DON'T USE! Kept for backwards-compatibility but should never have been public.
 def build_extra_kwargs(
     extra_kwargs: dict[str, Any],
     values: dict[str, Any],

libs/core/tests/unit_tests/utils/test_imports.py

@@ -17,8 +17,8 @@ EXPECTED_ALL = [
     "raise_for_status_with_text",
     "xor_args",
     "try_load_from_hub",
-    "build_extra_kwargs",
     "image",
+    "build_extra_kwargs",
     "get_from_dict_or_env",
     "get_from_env",
     "stringify_dict",

libs/partners/anthropic/langchain_anthropic/chat_models.py

@@ -56,13 +56,13 @@ from langchain_core.runnables import (
 )
 from langchain_core.tools import BaseTool
 from langchain_core.utils import (
-    build_extra_kwargs,
     from_env,
     get_pydantic_field_names,
     secret_from_env,
 )
 from langchain_core.utils.function_calling import convert_to_openai_tool
 from langchain_core.utils.pydantic import is_basemodel_subclass
+from langchain_core.utils.utils import _build_model_kwargs
 from pydantic import (
     BaseModel,
     ConfigDict,
@@ -646,11 +646,8 @@ class ChatAnthropic(BaseChatModel):
     @model_validator(mode="before")
     @classmethod
     def build_extra(cls, values: Dict) -> Any:
-        extra = values.get("model_kwargs", {})
         all_required_field_names = get_pydantic_field_names(cls)
-        values["model_kwargs"] = build_extra_kwargs(
-            extra, values, all_required_field_names
-        )
+        values = _build_model_kwargs(values, all_required_field_names)
         return values
 
     @model_validator(mode="after")

libs/partners/anthropic/langchain_anthropic/llms.py

@@ -25,7 +25,7 @@ from langchain_core.utils import (
     get_pydantic_field_names,
 )
 from langchain_core.utils.utils import (
-    build_extra_kwargs,
+    _build_model_kwargs,
     from_env,
     secret_from_env,
 )
@@ -88,11 +88,8 @@ class _AnthropicCommon(BaseLanguageModel):
     @model_validator(mode="before")
     @classmethod
     def build_extra(cls, values: Dict) -> Any:
-        extra = values.get("model_kwargs", {})
         all_required_field_names = get_pydantic_field_names(cls)
-        values["model_kwargs"] = build_extra_kwargs(
-            extra, values, all_required_field_names
-        )
+        values = _build_model_kwargs(values, all_required_field_names)
         return values
 
     @model_validator(mode="after")

libs/partners/anthropic/tests/unit_tests/test_chat_models.py

@@ -61,9 +61,12 @@ def test_anthropic_model_kwargs() -> None:
 
 
 @pytest.mark.requires("anthropic")
-def test_anthropic_invalid_model_kwargs() -> None:
-    with pytest.raises(ValueError):
-        ChatAnthropic(model="foo", model_kwargs={"max_tokens_to_sample": 5})  # type: ignore[call-arg]
+def test_anthropic_fields_in_model_kwargs() -> None:
+    """Test that for backwards compatibility fields can be passed in as model_kwargs."""
+    llm = ChatAnthropic(model="foo", model_kwargs={"max_tokens_to_sample": 5})  # type: ignore[call-arg]
+    assert llm.max_tokens == 5
+    llm = ChatAnthropic(model="foo", model_kwargs={"max_tokens": 5})  # type: ignore[call-arg]
+    assert llm.max_tokens == 5
 
 
 @pytest.mark.requires("anthropic")

libs/partners/fireworks/langchain_fireworks/chat_models.py

@@ -79,7 +79,7 @@ from langchain_core.utils.function_calling import (
     convert_to_openai_tool,
 )
 from langchain_core.utils.pydantic import is_basemodel_subclass
-from langchain_core.utils.utils import build_extra_kwargs, from_env, secret_from_env
+from langchain_core.utils.utils import _build_model_kwargs, from_env, secret_from_env
 from pydantic import (
     BaseModel,
     ConfigDict,
@@ -366,10 +366,7 @@ class ChatFireworks(BaseChatModel):
     def build_extra(cls, values: Dict[str, Any]) -> Any:
         """Build extra kwargs from additional params that were passed in."""
         all_required_field_names = get_pydantic_field_names(cls)
-        extra = values.get("model_kwargs", {})
-        values["model_kwargs"] = build_extra_kwargs(
-            extra, values, all_required_field_names
-        )
+        values = _build_model_kwargs(values, all_required_field_names)
         return values
 
     @model_validator(mode="after")

libs/partners/fireworks/langchain_fireworks/llms.py

@@ -11,7 +11,7 @@ from langchain_core.callbacks import (
 )
 from langchain_core.language_models.llms import LLM
 from langchain_core.utils import get_pydantic_field_names
-from langchain_core.utils.utils import build_extra_kwargs, secret_from_env
+from langchain_core.utils.utils import _build_model_kwargs, secret_from_env
 from pydantic import ConfigDict, Field, SecretStr, model_validator
 
 from langchain_fireworks.version import __version__
@@ -93,10 +93,7 @@ class Fireworks(LLM):
     def build_extra(cls, values: Dict[str, Any]) -> Any:
         """Build extra kwargs from additional params that were passed in."""
         all_required_field_names = get_pydantic_field_names(cls)
-        extra = values.get("model_kwargs", {})
-        values["model_kwargs"] = build_extra_kwargs(
-            extra, values, all_required_field_names
-        )
+        values = _build_model_kwargs(values, all_required_field_names)
         return values
 
     @property

libs/partners/openai/langchain_openai/chat_models/base.py

@@ -90,7 +90,7 @@ from langchain_core.utils.pydantic import (
     TypeBaseModel,
     is_basemodel_subclass,
 )
-from langchain_core.utils.utils import build_extra_kwargs, from_env, secret_from_env
+from langchain_core.utils.utils import _build_model_kwargs, from_env, secret_from_env
 from pydantic import BaseModel, ConfigDict, Field, SecretStr, model_validator
 from typing_extensions import Self
 
@@ -477,10 +477,7 @@ class BaseChatOpenAI(BaseChatModel):
     def build_extra(cls, values: Dict[str, Any]) -> Any:
         """Build extra kwargs from additional params that were passed in."""
         all_required_field_names = get_pydantic_field_names(cls)
-        extra = values.get("model_kwargs", {})
-        values["model_kwargs"] = build_extra_kwargs(
-            extra, values, all_required_field_names
-        )
+        values = _build_model_kwargs(values, all_required_field_names)
         return values
 
     @model_validator(mode="after")

libs/partners/openai/langchain_openai/llms/base.py

@@ -27,7 +27,7 @@ from langchain_core.callbacks import (
 from langchain_core.language_models.llms import BaseLLM
 from langchain_core.outputs import Generation, GenerationChunk, LLMResult
 from langchain_core.utils import get_pydantic_field_names
-from langchain_core.utils.utils import build_extra_kwargs, from_env, secret_from_env
+from langchain_core.utils.utils import _build_model_kwargs, from_env, secret_from_env
 from pydantic import ConfigDict, Field, SecretStr, model_validator
 from typing_extensions import Self
 
@@ -160,10 +160,7 @@ class BaseOpenAI(BaseLLM):
     def build_extra(cls, values: Dict[str, Any]) -> Any:
         """Build extra kwargs from additional params that were passed in."""
         all_required_field_names = get_pydantic_field_names(cls)
-        extra = values.get("model_kwargs", {})
-        values["model_kwargs"] = build_extra_kwargs(
-            extra, values, all_required_field_names
-        )
+        values = _build_model_kwargs(values, all_required_field_names)
         return values
 
     @model_validator(mode="after")

libs/partners/openai/tests/unit_tests/llms/test_base.py

@@ -30,9 +30,12 @@ def test_openai_model_kwargs() -> None:
     assert llm.model_kwargs == {"foo": "bar"}
 
 
-def test_openai_invalid_model_kwargs() -> None:
-    with pytest.raises(ValueError):
-        OpenAI(model_kwargs={"model_name": "foo"})
+def test_openai_fields_in_model_kwargs() -> None:
+    """Test that for backwards compatibility fields can be passed in as model_kwargs."""
+    llm = OpenAI(model_kwargs={"model_name": "foo"})
+    assert llm.model_name == "foo"
+    llm = OpenAI(model_kwargs={"model": "foo"})
+    assert llm.model_name == "foo"
 
 
 def test_openai_incorrect_field() -> None: