use LangChain with structured output

Author: ks6088ts

apps/1_call_azure_openai_chat/convert_code.py

@@ -3,8 +3,10 @@
 from os import getenv
 
 from dotenv import load_dotenv
+from langchain_core.messages import HumanMessage, SystemMessage
+from langchain_openai import AzureChatOpenAI
 from openai import AzureOpenAI
-from pydantic import BaseModel
+from pydantic import BaseModel, Field
 
 
 def init_args() -> argparse.Namespace:
@@ -14,18 +16,19 @@ def init_args() -> argparse.Namespace:
     )
     parser.add_argument("-s", "--system", default="Extract the event information.")
     parser.add_argument("-u", "--user", default="Alice and Bob are going to a science fair on Friday.")
+    parser.add_argument("-t", "--type", default="openai", choices=["openai", "langchain"])
     parser.add_argument("-v", "--verbose", action="store_true")
     return parser.parse_args()
 
 
 class ResponseFormat(BaseModel):
     # FIXME: Update the ResponseFormat class to match your actual response format
-    name: str
-    date: str
-    participants: list[str]
+    name: str = Field(description="The name of the event.")
+    date: str = Field(description="The date of the event.")
+    participants: list[str] = Field(description="The participants of the event.")
 
 
-def print_structured_output(
+def print_structured_output_openai(
     system: str,
     user: str,
 ):
@@ -60,6 +63,43 @@ def print_structured_output(
     print(completion.model_dump_json(indent=2))
 
 
+def print_structured_output_langchain(
+    system: str,
+    user: str,
+):
+    """
+    How to use:
+        Support for structured outputs was first added in API version 2024-08-01-preview.
+        It is available in the latest preview APIs as well as the latest GA API: 2024-10-21.
+
+    Install dependencies:
+        $ pip install openai python-dotenv pydantic langchain-openai
+
+    References:
+        - https://python.langchain.com/docs/how_to/structured_output/
+    """
+    llm = AzureChatOpenAI(
+        temperature=0,
+        api_key=getenv("AZURE_OPENAI_API_KEY"),
+        api_version=getenv("AZURE_OPENAI_API_VERSION"),
+        azure_endpoint=getenv("AZURE_OPENAI_ENDPOINT"),
+        model=getenv("AZURE_OPENAI_GPT_MODEL"),
+    )
+    structured_llm = llm.with_structured_output(ResponseFormat)
+    response: ResponseFormat = structured_llm.invoke(
+        [
+            SystemMessage(
+                content=system,
+            ),
+            HumanMessage(
+                content=user,
+            ),
+        ]
+    )
+    print(response)
+    print(f"Name: {response.name}, Date: {response.date}, Participants: {response.participants}")
+
+
 if __name__ == "__main__":
     args = init_args()
 
@@ -71,9 +111,18 @@ def print_structured_output(
     load_dotenv()
 
     try:
-        print_structured_output(
-            system=args.system,
-            user=args.user,
-        )
+        if args.type == "openai":
+            print_structured_output_openai(
+                system=args.system,
+                user=args.user,
+            )
+        elif args.type == "langchain":
+            print_structured_output_langchain(
+                system=args.system,
+                user=args.user,
+            )
+        else:
+            raise ValueError(f"Invalid type: {args.type}")
+
     except Exception as e:
         logging.error(e)