feat(docs): clarify structured output with tools ordering (#32527)

This commit is contained in:
Mason Daugherty 2025-08-13 10:40:48 -04:00 committed by GitHub
parent 024422e9b0
commit 7932e1edd1
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 153 additions and 0 deletions

View File

@ -29,6 +29,22 @@ model_with_structure = model.with_structured_output(schema)
structured_output = model_with_structure.invoke(user_input) structured_output = model_with_structure.invoke(user_input)
``` ```
:::warning[Tool Order Matters]
When combining structured output with additional tools, bind tools **first**, then apply structured output:
```python
# Correct
model_with_tools = model.bind_tools([tool1, tool2])
structured_model = model_with_tools.with_structured_output(schema)
# Incorrect - will cause tool resolution errors
structured_model = model.with_structured_output(schema)
broken_model = structured_model.bind_tools([tool1, tool2])
```
:::
## Schema definition ## Schema definition
The central concept is that the output structure of model responses needs to be represented in some way. The central concept is that the output structure of model responses needs to be represented in some way.

View File

@ -998,6 +998,91 @@
"\n", "\n",
"chain.invoke({\"query\": query})" "chain.invoke({\"query\": query})"
] ]
},
{
"cell_type": "markdown",
"id": "xfejabhtn2",
"metadata": {},
"source": [
"## Combining with Additional Tools\n",
"\n",
"When you need to use both structured output and additional tools (like web search), note the order of operations:\n",
"\n",
"**Correct Order**:\n",
"```python\n",
"# 1. Bind tools first\n",
"llm_with_tools = llm.bind_tools([web_search_tool, calculator_tool])\n",
"\n",
"# 2. Apply structured output\n",
"structured_llm = llm_with_tools.with_structured_output(MySchema)\n",
"```\n",
"\n",
"**Incorrect Order**:\n",
"\n",
"```python\n",
"# This will fail with \"Tool 'MySchema' not found\" error\n",
"structured_llm = llm.with_structured_output(MySchema)\n",
"broken_llm = structured_llm.bind_tools([web_search_tool])\n",
"```"
]
},
{
"cell_type": "markdown",
"id": "653798ca",
"metadata": {},
"source": [
"**Why Order Matters:**\n",
"`with_structured_output()` internally uses tool calling to enforce the schema. When you bind additional tools afterward, it creates a conflict in the tool resolution system."
]
},
{
"cell_type": "markdown",
"id": "1345f4a4",
"metadata": {},
"source": [
"**Complete Example:**"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "0835637b",
"metadata": {},
"outputs": [],
"source": [
"from pydantic import BaseModel, Field\n",
"from langchain_openai import ChatOpenAI\n",
"\n",
"\n",
"class SearchResult(BaseModel):\n",
" \"\"\"Structured search result.\"\"\"\n",
"\n",
" query: str = Field(description=\"The search query\")\n",
" findings: str = Field(description=\"Summary of findings\")\n",
"\n",
"\n",
"# Define tools\n",
"search_tool = {\n",
" \"type\": \"function\",\n",
" \"function\": {\n",
" \"name\": \"web_search\",\n",
" \"description\": \"Search the web for information\",\n",
" \"parameters\": {\n",
" \"type\": \"object\",\n",
" \"properties\": {\"query\": {\"type\": \"string\", \"description\": \"Search query\"}},\n",
" \"required\": [\"query\"],\n",
" },\n",
" },\n",
"}\n",
"\n",
"# Correct approach\n",
"llm = ChatOpenAI()\n",
"llm_with_search = llm.bind_tools([search_tool])\n",
"structured_search_llm = llm_with_search.with_structured_output(SearchResult)\n",
"\n",
"# Now you can use both search and get structured output\n",
"result = structured_search_llm.invoke(\"Search for latest AI research and summarize\")"
]
} }
], ],
"metadata": { "metadata": {

View File

@ -1240,6 +1240,58 @@
"response = llm_with_tools.invoke(\"How do I update a web app to TypeScript 5.5?\")" "response = llm_with_tools.invoke(\"How do I update a web app to TypeScript 5.5?\")"
] ]
}, },
{
"cell_type": "markdown",
"id": "kloc4rvd1w",
"metadata": {},
"source": [
"#### Web search + structured output\n",
"\n",
"When combining web search tools with structured output, it's important to **bind the tools first and then apply structured output**:"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "rjjergy6ef",
"metadata": {},
"outputs": [],
"source": [
"from pydantic import BaseModel, Field\n",
"from langchain_anthropic import ChatAnthropic\n",
"\n",
"\n",
"# Define structured output schema\n",
"class ResearchResult(BaseModel):\n",
" \"\"\"Structured research result from web search.\"\"\"\n",
"\n",
" topic: str = Field(description=\"The research topic\")\n",
" summary: str = Field(description=\"Summary of key findings\")\n",
" key_points: list[str] = Field(description=\"List of important points discovered\")\n",
"\n",
"\n",
"# Configure web search tool\n",
"websearch_tools = [\n",
" {\n",
" \"type\": \"web_search_20250305\",\n",
" \"name\": \"web_search\",\n",
" \"max_uses\": 10,\n",
" }\n",
"]\n",
"\n",
"llm = ChatAnthropic(model=\"claude-3-5-sonnet-20241022\")\n",
"\n",
"# Correct order: bind tools first, then structured output\n",
"llm_with_search = llm.bind_tools(websearch_tools)\n",
"research_llm = llm_with_search.with_structured_output(ResearchResult)\n",
"\n",
"# Now you can use both web search and get structured output\n",
"result = research_llm.invoke(\"Research the latest developments in quantum computing\")\n",
"print(f\"Topic: {result.topic}\")\n",
"print(f\"Summary: {result.summary}\")\n",
"print(f\"Key Points: {result.key_points}\")"
]
},
{ {
"cell_type": "markdown", "cell_type": "markdown",
"id": "1478cdc6-2e52-4870-80f9-b4ddf88f2db2", "id": "1478cdc6-2e52-4870-80f9-b4ddf88f2db2",