openai[patch]: get output_type when using with_structured_output (#26307)

- This allows pydantic to correctly resolve annotations necessary when
using openai new param `json_schema`

Resolves issue: #26250

---------

Co-authored-by: Eugene Yurtsev <[email protected]>
Co-authored-by: Bagatur <[email protected]>

Author: 3 people

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

@@ -65,7 +65,6 @@
 from langchain_core.messages.ai import UsageMetadata
 from langchain_core.messages.tool import tool_call_chunk
 from langchain_core.output_parsers import JsonOutputParser, PydanticOutputParser
-from langchain_core.output_parsers.base import OutputParserLike
 from langchain_core.output_parsers.openai_tools import (
     JsonOutputKeyToolsParser,
     PydanticToolsParser,
@@ -1421,7 +1420,7 @@ class AnswerWithJustification(BaseModel):
                 strict=strict,
             )
             if is_pydantic_schema:
-                output_parser: OutputParserLike = PydanticToolsParser(
+                output_parser: Runnable = PydanticToolsParser(
                     tools=[schema],  # type: ignore[list-item]
                     first_tool_only=True,  # type: ignore[list-item]
                 )
@@ -1445,11 +1444,12 @@ class AnswerWithJustification(BaseModel):
             strict = strict if strict is not None else True
             response_format = _convert_to_openai_response_format(schema, strict=strict)
             llm = self.bind(response_format=response_format)
-            output_parser = (
-                cast(Runnable, _oai_structured_outputs_parser)
-                if is_pydantic_schema
-                else JsonOutputParser()
-            )
+            if is_pydantic_schema:
+                output_parser = _oai_structured_outputs_parser.with_types(
+                    output_type=cast(type, schema)
+                )
+            else:
+                output_parser = JsonOutputParser()
         else:
             raise ValueError(
                 f"Unrecognized method argument. Expected one of 'function_calling' or "

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

@@ -18,6 +18,7 @@
 )
 from langchain_core.messages.ai import UsageMetadata
 from langchain_core.pydantic_v1 import BaseModel
+from pydantic import BaseModel as BaseModelV2
 
 from langchain_openai import ChatOpenAI
 from langchain_openai.chat_models.base import (
@@ -694,3 +695,31 @@ def test_get_num_tokens_from_messages() -> None:
     expected = 176
     actual = llm.get_num_tokens_from_messages(messages)
     assert expected == actual
+
+
+class Foo(BaseModel):
+    bar: int
+
+
+class FooV2(BaseModelV2):
+    bar: int
+
+
+@pytest.mark.parametrize("schema", [Foo, FooV2])
+def test_schema_from_with_structured_output(schema: Type) -> None:
+    """Test schema from with_structured_output."""
+
+    llm = ChatOpenAI()
+
+    structured_llm = llm.with_structured_output(
+        schema, method="json_schema", strict=True
+    )
+
+    expected = {
+        "properties": {"bar": {"title": "Bar", "type": "integer"}},
+        "required": ["bar"],
+        "title": schema.__name__,
+        "type": "object",
+    }
+    actual = structured_llm.get_output_schema().schema()
+    assert actual == expected