github/langchain
1
0
Fork 0
mirror of https://github.com/hwchase17/langchain.git synced 2025-09-18 16:16:33 +00:00
Code Issues Packages Projects Releases Wiki Activity

ollama[patch]: support structured output (#28629)

Browse Source 
- Bump minimum version of `ollama` to 0.4.4 (which also addresses
https://github.com/langchain-ai/langchain/issues/28607).
- Support recently-released [structured
output](https://ollama.com/blog/structured-outputs) feature. This can be
accessed by calling `.with_structured_output` with
`method="json_schema"` (choice of name
[mirrors](https://python.langchain.com/api_reference/openai/chat_models/langchain_openai.chat_models.base.ChatOpenAI.html#langchain_openai.chat_models.base.ChatOpenAI.with_structured_output)
what we have for OpenAI's structured output feature).

`ChatOllama` previously implemented `.with_structured_output` via the
[base
implementation](ec9b41431e/libs/core/langchain_core/language_models/chat_models.py (L1117)).
This commit is contained in:
ccurme
2024-12-10 10:36:00 -05:00
committed by GitHub
parent 24292c4a31
commit 5c6e2cbcda
4 changed files with 642 additions and 91 deletions
Download Patch File Download Diff File Expand all files Collapse all files

345
libs/partners/ollama/langchain_ollama/chat_models.py
View File

@@ -1,6 +1,7 @@
"""Ollama chat models."""
import json
from operator import itemgetter
from typing import (
Any,
AsyncIterator,
@@ -36,13 +37,24 @@ from langchain_core.messages import (
)
from langchain_core.messages.ai import UsageMetadata
from langchain_core.messages.tool import tool_call
from langchain_core.output_parsers import (
JsonOutputKeyToolsParser,
JsonOutputParser,
PydanticOutputParser,
PydanticToolsParser,
)
from langchain_core.outputs import ChatGeneration, ChatGenerationChunk, ChatResult
from langchain_core.runnables import Runnable
from langchain_core.runnables import Runnable, RunnableMap, RunnablePassthrough
from langchain_core.tools import BaseTool
from langchain_core.utils.function_calling import (
_convert_any_typed_dicts_to_pydantic as convert_any_typed_dicts_to_pydantic,
)
from langchain_core.utils.function_calling import convert_to_openai_tool
from langchain_core.utils.pydantic import TypeBaseModel, is_basemodel_subclass
from ollama import AsyncClient, Client, Message, Options
from pydantic import PrivateAttr, model_validator
from typing_extensions import Self
from pydantic import BaseModel, PrivateAttr, model_validator
from pydantic.json_schema import JsonSchemaValue
from typing_extensions import Self, is_typeddict
def _get_usage_metadata_from_generation_info(
@@ -157,6 +169,10 @@ def _lc_tool_call_to_openai_tool_call(tool_call: ToolCall) -> dict:
}
def _is_pydantic_class(obj: Any) -> bool:
return isinstance(obj, type) and is_basemodel_subclass(obj)
class ChatOllama(BaseChatModel):
r"""Ollama chat model integration.
@@ -385,8 +401,8 @@ class ChatOllama(BaseChatModel):
to more diverse text, while a lower value (e.g., 0.5) will
generate more focused and conservative text. (Default: 0.9)"""
format: Literal["", "json"] = ""
"""Specify the format of the output (options: json)"""
format: Optional[Union[Literal["", "json"], JsonSchemaValue]] = None
"""Specify the format of the output (options: "json", JSON schema)."""
keep_alive: Optional[Union[int, str]] = None
"""How long the model will stay loaded into memory."""
@@ -815,3 +831,322 @@ class ChatOllama(BaseChatModel):
""" # noqa: E501
formatted_tools = [convert_to_openai_tool(tool) for tool in tools]
return super().bind(tools=formatted_tools, **kwargs)
def with_structured_output(
self,
schema: Union[Dict, type],
*,
method: Literal[
"function_calling", "json_mode", "json_schema"
] = "function_calling",
include_raw: bool = False,
**kwargs: Any,
) -> Runnable[LanguageModelInput, Union[Dict, BaseModel]]:
"""Model wrapper that returns outputs formatted to match the given schema.
Args:
schema:
The output schema. Can be passed in as:
- a Pydantic class,
- a JSON schema
- a TypedDict class
- an OpenAI function/tool schema.
If ``schema`` is a Pydantic class then the model output will be a
Pydantic instance of that class, and the model-generated fields will be
validated by the Pydantic class. Otherwise the model output will be a
dict and will not be validated. See :meth:`langchain_core.utils.function_calling.convert_to_openai_tool`
for more on how to properly specify types and descriptions of
schema fields when specifying a Pydantic or TypedDict class.
method: The method for steering model generation, one of:
- "function_calling":
Uses Ollama's tool-calling API
- "json_schema":
Uses Ollama's structured output API: https://ollama.com/blog/structured-outputs
- "json_mode":
Specifies ``format="json"``. Note that if using JSON mode then you
must include instructions for formatting the output into the
desired schema into the model call.
include_raw:
If False then only the parsed structured output is returned. If
an error occurs during model output parsing it will be raised. If True
then both the raw model response (a BaseMessage) and the parsed model
response will be returned. If an error occurs during output parsing it
will be caught and returned as well. The final output is always a dict
with keys "raw", "parsed", and "parsing_error".
kwargs: Additional keyword args aren't supported.
Returns:
A Runnable that takes same inputs as a :class:`langchain_core.language_models.chat.BaseChatModel`.
| If ``include_raw`` is False and ``schema`` is a Pydantic class, Runnable outputs an instance of ``schema`` (i.e., a Pydantic object). Otherwise, if ``include_raw`` is False then Runnable outputs a dict.
| If ``include_raw`` is True, then Runnable outputs a dict with keys:
- "raw": BaseMessage
- "parsed": None if there was a parsing error, otherwise the type depends on the ``schema`` as described above.
- "parsing_error": Optional[BaseException]
.. versionchanged:: 0.2.2
Added support for structured output API via ``format`` parameter.
.. dropdown:: Example: schema=Pydantic class, method="function_calling", include_raw=False
.. code-block:: python
from typing import Optional
from langchain_ollama import ChatOllama
from pydantic import BaseModel, Field
class AnswerWithJustification(BaseModel):
'''An answer to the user question along with justification for the answer.'''
answer: str
justification: Optional[str] = Field(
default=..., description="A justification for the answer."
)
llm = ChatOllama(model="llama3.1", temperature=0)
structured_llm = llm.with_structured_output(
AnswerWithJustification
)
structured_llm.invoke(
"What weighs more a pound of bricks or a pound of feathers"
)
# -> AnswerWithJustification(
# answer='They weigh the same',
# justification='Both a pound of bricks and a pound of feathers weigh one pound. The weight is the same, but the volume or density of the objects may differ.'
# )
.. dropdown:: Example: schema=Pydantic class, method="function_calling", include_raw=True
.. code-block:: python
from langchain_ollama import ChatOllama
from pydantic import BaseModel
class AnswerWithJustification(BaseModel):
'''An answer to the user question along with justification for the answer.'''
answer: str
justification: str
llm = ChatOllama(model="llama3.1", temperature=0)
structured_llm = llm.with_structured_output(
AnswerWithJustification, include_raw=True
)
structured_llm.invoke(
"What weighs more a pound of bricks or a pound of feathers"
)
# -> {
# 'raw': AIMessage(content='', additional_kwargs={'tool_calls': [{'id': 'call_Ao02pnFYXD6GN1yzc0uXPsvF', 'function': {'arguments': '{"answer":"They weigh the same.","justification":"Both a pound of bricks and a pound of feathers weigh one pound. The weight is the same, but the volume or density of the objects may differ."}', 'name': 'AnswerWithJustification'}, 'type': 'function'}]}),
# 'parsed': AnswerWithJustification(answer='They weigh the same.', justification='Both a pound of bricks and a pound of feathers weigh one pound. The weight is the same, but the volume or density of the objects may differ.'),
# 'parsing_error': None
# }
.. dropdown:: Example: schema=Pydantic class, method="json_schema", include_raw=False
.. code-block:: python
from typing import Optional
from langchain_ollama import ChatOllama
from pydantic import BaseModel, Field
class AnswerWithJustification(BaseModel):
'''An answer to the user question along with justification for the answer.'''
answer: str
justification: Optional[str] = Field(
default=..., description="A justification for the answer."
)
llm = ChatOllama(model="llama3.1", temperature=0)
structured_llm = llm.with_structured_output(
AnswerWithJustification, method="json_schema"
)
structured_llm.invoke(
"What weighs more a pound of bricks or a pound of feathers"
)
# -> AnswerWithJustification(
# answer='They weigh the same',
# justification='Both a pound of bricks and a pound of feathers weigh one pound. The weight is the same, but the volume or density of the objects may differ.'
# )
.. dropdown:: Example: schema=TypedDict class, method="function_calling", include_raw=False
.. code-block:: python
# IMPORTANT: If you are using Python <=3.8, you need to import Annotated
# from typing_extensions, not from typing.
from typing_extensions import Annotated, TypedDict
from langchain_ollama import ChatOllama
class AnswerWithJustification(TypedDict):
'''An answer to the user question along with justification for the answer.'''
answer: str
justification: Annotated[
Optional[str], None, "A justification for the answer."
]
llm = ChatOllama(model="llama3.1", temperature=0)
structured_llm = llm.with_structured_output(AnswerWithJustification)
structured_llm.invoke(
"What weighs more a pound of bricks or a pound of feathers"
)
# -> {
# 'answer': 'They weigh the same',
# 'justification': 'Both a pound of bricks and a pound of feathers weigh one pound. The weight is the same, but the volume and density of the two substances differ.'
# }
.. dropdown:: Example: schema=OpenAI function schema, method="function_calling", include_raw=False
.. code-block:: python
from langchain_ollama import ChatOllama
oai_schema = {
'name': 'AnswerWithJustification',
'description': 'An answer to the user question along with justification for the answer.',
'parameters': {
'type': 'object',
'properties': {
'answer': {'type': 'string'},
'justification': {'description': 'A justification for the answer.', 'type': 'string'}
},
'required': ['answer']
}
}
llm = ChatOllama(model="llama3.1", temperature=0)
structured_llm = llm.with_structured_output(oai_schema)
structured_llm.invoke(
"What weighs more a pound of bricks or a pound of feathers"
)
# -> {
# 'answer': 'They weigh the same',
# 'justification': 'Both a pound of bricks and a pound of feathers weigh one pound. The weight is the same, but the volume and density of the two substances differ.'
# }
.. dropdown:: Example: schema=Pydantic class, method="json_mode", include_raw=True
.. code-block::
from langchain_ollama import ChatOllama
from pydantic import BaseModel
class AnswerWithJustification(BaseModel):
answer: str
justification: str
llm = ChatOllama(model="llama3.1", temperature=0)
structured_llm = llm.with_structured_output(
AnswerWithJustification,
method="json_mode",
include_raw=True
)
structured_llm.invoke(
"Answer the following question. "
"Make sure to return a JSON blob with keys 'answer' and 'justification'.\\n\\n"
"What's heavier a pound of bricks or a pound of feathers?"
)
# -> {
# 'raw': AIMessage(content='{\\n "answer": "They are both the same weight.",\\n "justification": "Both a pound of bricks and a pound of feathers weigh one pound. The difference lies in the volume and density of the materials, not the weight." \\n}'),
# 'parsed': AnswerWithJustification(answer='They are both the same weight.', justification='Both a pound of bricks and a pound of feathers weigh one pound. The difference lies in the volume and density of the materials, not the weight.'),
# 'parsing_error': None
# }
""" # noqa: E501, D301
if kwargs:
raise ValueError(f"Received unsupported arguments {kwargs}")
is_pydantic_schema = _is_pydantic_class(schema)
if method == "function_calling":
if schema is None:
raise ValueError(
"schema must be specified when method is not 'json_mode'. "
"Received None."
)
tool_name = convert_to_openai_tool(schema)["function"]["name"]
llm = self.bind_tools([schema], tool_choice=tool_name)
if is_pydantic_schema:
output_parser: Runnable = PydanticToolsParser(
tools=[schema], # type: ignore[list-item]
first_tool_only=True,
)
else:
output_parser = JsonOutputKeyToolsParser(
key_name=tool_name, first_tool_only=True
)
elif method == "json_mode":
llm = self.bind(format="json")
output_parser = (
PydanticOutputParser(pydantic_object=schema) # type: ignore[arg-type]
if is_pydantic_schema
else JsonOutputParser()
)
elif method == "json_schema":
if schema is None:
raise ValueError(
"schema must be specified when method is not 'json_mode'. "
"Received None."
)
if is_pydantic_schema:
schema = cast(TypeBaseModel, schema)
llm = self.bind(format=schema.model_json_schema())
output_parser = PydanticOutputParser(pydantic_object=schema)
else:
if is_typeddict(schema):
schema = cast(type, schema)
response_format = convert_any_typed_dicts_to_pydantic(
schema, visited={}
).schema() # type: ignore[attr-defined]
if "required" not in response_format:
response_format["required"] = list(
response_format["properties"].keys()
)
else:
# is JSON schema
response_format = schema
llm = self.bind(format=response_format)
output_parser = JsonOutputParser()
else:
raise ValueError(
f"Unrecognized method argument. Expected one of 'function_calling', "
f"'json_schema', or 'json_mode'. Received: '{method}'"
)
if include_raw:
parser_assign = RunnablePassthrough.assign(
parsed=itemgetter("raw") | output_parser, parsing_error=lambda _: None
)
parser_none = RunnablePassthrough.assign(parsed=lambda _: None)
parser_with_fallback = parser_assign.with_fallbacks(
[parser_none], exception_key="parsing_error"
)
return RunnableMap(raw=llm) | parser_with_fallback
else:
return llm | output_parser