langchain-ollama 0.3.3__py3-none-any.whl → 0.3.4__py3-none-any.whl

langchain_ollama/__init__.py

@@ -18,7 +18,7 @@ del metadata  # optional, avoids polluting the results of dir(__package__)
 
 __all__ = [
     "ChatOllama",
-    "OllamaLLM",
     "OllamaEmbeddings",
+    "OllamaLLM",
     "__version__",
 ]

langchain_ollama/_utils.py

@@ -0,0 +1,39 @@
+"""Utility functions for validating Ollama models."""
+
+from httpx import ConnectError
+from ollama import Client, ResponseError
+
+
+def validate_model(client: Client, model_name: str) -> None:
+    """Validate that a model exists in the Ollama instance.
+
+    Args:
+        client: The Ollama client.
+        model_name: The name of the model to validate.
+
+    Raises:
+        ValueError: If the model is not found or if there's a connection issue.
+    """
+    try:
+        response = client.list()
+
+        model_names: list[str] = [model["model"] for model in response["models"]]
+
+        if not any(
+            model_name == m or m.startswith(f"{model_name}:") for m in model_names
+        ):
+            msg = (
+                f"Model `{model_name}` not found in Ollama. Please pull the "
+                f"model (using `ollama pull {model_name}`) or specify a valid "
+                f"model name. Available local models: {', '.join(model_names)}"
+            )
+            raise ValueError(msg)
+    except ConnectError as e:
+        msg = "Failed to connect to Ollama. Please check that Ollama is downloaded, running and accessible. https://ollama.com/download"  # noqa: E501
+        raise ValueError(msg) from e
+    except ResponseError as e:
+        msg = (
+            "Received an error from the Ollama API. "
+            "Please check your Ollama server logs."
+        )
+        raise ValueError(msg) from e

langchain_ollama/chat_models.py

@@ -1,12 +1,13 @@
 """Ollama chat models."""
 
+from __future__ import annotations
+
 import json
 from collections.abc import AsyncIterator, Iterator, Mapping, Sequence
 from operator import itemgetter
 from typing import (
     Any,
     Callable,
-    Final,
     Literal,
     Optional,
     Union,
@@ -25,7 +26,6 @@ from langchain_core.messages import (
     AIMessage,
     AIMessageChunk,
     BaseMessage,
-    BaseMessageChunk,
     ChatMessage,
     HumanMessage,
     SystemMessage,
@@ -55,8 +55,7 @@ from pydantic.json_schema import JsonSchemaValue
 from pydantic.v1 import BaseModel as BaseModelV1
 from typing_extensions import Self, is_typeddict
 
-DEFAULT_THINK_TOKEN_START: Final[str] = "<think>"
-DEFAULT_THINK_TOKEN_END: Final[str] = "</think>"
+from ._utils import validate_model
 
 
 def _get_usage_metadata_from_generation_info(
@@ -77,7 +76,9 @@ def _get_usage_metadata_from_generation_info(
 
 
 def _parse_json_string(
-    json_string: str, raw_tool_call: dict[str, Any], skip: bool
+    json_string: str,
+    raw_tool_call: dict[str, Any],
+    skip: bool,  # noqa: FBT001
 ) -> Any:
     """Attempt to parse a JSON string for tool calling.
 
@@ -151,26 +152,30 @@ def _get_tool_calls_from_response(
 ) -> list[ToolCall]:
     """Get tool calls from ollama response."""
     tool_calls = []
-    if "message" in response:
-        if raw_tool_calls := response["message"].get("tool_calls"):
-            for tc in raw_tool_calls:
-                tool_calls.append(
-                    tool_call(
-                        id=str(uuid4()),
-                        name=tc["function"]["name"],
-                        args=_parse_arguments_from_tool_call(tc) or {},
-                    )
+    if "message" in response and (
+        raw_tool_calls := response["message"].get("tool_calls")
+    ):
+        tool_calls.extend(
+            [
+                tool_call(
+                    id=str(uuid4()),
+                    name=tc["function"]["name"],
+                    args=_parse_arguments_from_tool_call(tc) or {},
                 )
+                for tc in raw_tool_calls
+            ]
+        )
     return tool_calls
 
 
-def _lc_tool_call_to_openai_tool_call(tool_call: ToolCall) -> dict:
+def _lc_tool_call_to_openai_tool_call(tool_call_: ToolCall) -> dict:
+    """Convert a LangChain tool call to an OpenAI tool call format."""
     return {
         "type": "function",
-        "id": tool_call["id"],
+        "id": tool_call_["id"],
         "function": {
-            "name": tool_call["name"],
-            "arguments": tool_call["args"],
+            "name": tool_call_["name"],
+            "arguments": tool_call_["args"],
         },
     }
 
@@ -180,14 +185,12 @@ def _get_image_from_data_content_block(block: dict) -> str:
     if block["type"] == "image":
         if block["source_type"] == "base64":
             return block["data"]
-        else:
-            error_message = "Image data only supported through in-line base64 format."
-            raise ValueError(error_message)
-
-    else:
-        error_message = f"Blocks of type {block['type']} not supported."
+        error_message = "Image data only supported through in-line base64 format."
         raise ValueError(error_message)
 
+    error_message = f"Blocks of type {block['type']} not supported."
+    raise ValueError(error_message)
+
 
 def _is_pydantic_class(obj: Any) -> bool:
     return isinstance(obj, type) and is_basemodel_subclass(obj)
@@ -209,8 +212,22 @@ class ChatOllama(BaseChatModel):
     Key init args — completion params:
         model: str
             Name of Ollama model to use.
+        reasoning: Optional[bool]
+            Controls the reasoning/thinking mode for
+            `supported models <https://ollama.com/search?c=thinking>`__.
+
+            - ``True``: Enables reasoning mode. The model's reasoning process will be
+              captured and returned separately in the ``additional_kwargs`` of the
+              response message, under ``reasoning_content``. The main response
+              content will not include the reasoning tags.
+            - ``False``: Disables reasoning mode. The model will not perform any reasoning,
+              and the response will not include any reasoning content.
+            - ``None`` (Default): The model will use its default reasoning behavior. Note
+              however, if the model's default behavior *is* to perform reasoning, think tags
+              (``<think>`` and ``</think>``) will be present within the main response content
+              unless you set ``reasoning`` to ``True``.
         temperature: float
-            Sampling temperature. Ranges from 0.0 to 1.0.
+            Sampling temperature. Ranges from ``0.0`` to ``1.0``.
         num_predict: Optional[int]
             Max number of tokens to generate.
 
@@ -326,7 +343,6 @@ class ChatOllama(BaseChatModel):
             '{"location": "Pune, India", "time_of_day": "morning"}'
 
     Tool Calling:
-
         .. code-block:: python
 
             from langchain_ollama import ChatOllama
@@ -345,17 +361,70 @@ class ChatOllama(BaseChatModel):
             'args': {'a': 45, 'b': 67},
             'id': '420c3f3b-df10-4188-945f-eb3abdb40622',
             'type': 'tool_call'}]
-    """  # noqa: E501
+
+    Thinking / Reasoning:
+        You can enable reasoning mode for models that support it by setting
+        the ``reasoning`` parameter to ``True`` in either the constructor or
+        the ``invoke``/``stream`` methods. This will enable the model to think
+        through the problem and return the reasoning process separately in the
+        ``additional_kwargs`` of the response message, under ``reasoning_content``.
+
+        If ``reasoning`` is set to ``None``, the model will use its default reasoning
+        behavior, and any reasoning content will *not* be captured under the
+        ``reasoning_content`` key, but will be present within the main response content
+        as think tags (``<think>`` and ``</think>``).
+
+        .. note::
+            This feature is only available for `models that support reasoning <https://ollama.com/search?c=thinking>`__.
+
+        .. code-block:: python
+
+            from langchain_ollama import ChatOllama
+
+            llm = ChatOllama(
+                model = "deepseek-r1:8b",
+                reasoning= True,
+            )
+
+            user_message = HumanMessage(content="how many r in the word strawberry?")
+            messages: List[Any] = [user_message]
+            llm.invoke(messages)
+
+            # or, on an invocation basis:
+
+            llm.invoke(messages, reasoning=True)
+            # or llm.stream(messages, reasoning=True)
+
+            # If not provided, the invocation will default to the ChatOllama reasoning
+            # param provided (None by default).
+
+        .. code-block:: python
+
+            AIMessage(content='The word "strawberry" contains **three \'r\' letters**. Here\'s a breakdown for clarity:\n\n- The spelling of "strawberry" has two parts ... be 3.\n\nTo be thorough, let\'s confirm with an online source or common knowledge.\n\nI can recall that "strawberry" has: s-t-r-a-w-b-e-r-r-y — yes, three r\'s.\n\nPerhaps it\'s misspelled by some, but standard is correct.\n\nSo I think the response should be 3.\n'}, response_metadata={'model': 'deepseek-r1:8b', 'created_at': '2025-07-08T19:33:55.891269Z', 'done': True, 'done_reason': 'stop', 'total_duration': 98232561292, 'load_duration': 28036792, 'prompt_eval_count': 10, 'prompt_eval_duration': 40171834, 'eval_count': 3615, 'eval_duration': 98163832416, 'model_name': 'deepseek-r1:8b'}, id='run--18f8269f-6a35-4a7c-826d-b89d52c753b3-0', usage_metadata={'input_tokens': 10, 'output_tokens': 3615, 'total_tokens': 3625})
+
+
+    """  # noqa: E501, pylint: disable=line-too-long
 
     model: str
     """Model name to use."""
 
-    extract_reasoning: Optional[Union[bool, tuple[str, str]]] = False
-    """Whether to extract the reasoning tokens in think blocks.
-    Extracts `chunk.content` to `chunk.additional_kwargs.reasoning_content`.
-    If a tuple is supplied, they are assumed to be the (start, end) tokens.
-    If `extract_reasoning=True`, the tokens will default to (<think>, </think>).
-    """
+    reasoning: Optional[bool] = None
+    """Controls the reasoning/thinking mode for
+    `supported models <https://ollama.com/search?c=thinking>`__.
+
+    - ``True``: Enables reasoning mode. The model's reasoning process will be
+      captured and returned separately in the ``additional_kwargs`` of the
+      response message, under ``reasoning_content``. The main response
+      content will not include the reasoning tags.
+    - ``False``: Disables reasoning mode. The model will not perform any reasoning,
+      and the response will not include any reasoning content.
+    - ``None`` (Default): The model will use its default reasoning behavior. Note
+      however, if the model's default behavior *is* to perform reasoning, think tags
+      ()``<think>`` and ``</think>``) will be present within the main response content
+      unless you set ``reasoning`` to ``True``."""
+
+    validate_model_on_init: bool = False
+    """Whether to validate the model exists in Ollama locally on initialization."""
 
     mirostat: Optional[int] = None
     """Enable Mirostat sampling for controlling perplexity.
@@ -436,7 +505,7 @@ class ChatOllama(BaseChatModel):
     """Base url the model is hosted under."""
 
     client_kwargs: Optional[dict] = {}
-    """Additional kwargs to pass to the httpx clients. 
+    """Additional kwargs to pass to the httpx clients.
     These arguments are passed to both synchronous and async clients.
     Use sync_client_kwargs and async_client_kwargs to pass different arguments
     to synchronous and asynchronous clients.
@@ -445,21 +514,21 @@ class ChatOllama(BaseChatModel):
     async_client_kwargs: Optional[dict] = {}
     """Additional kwargs to merge with client_kwargs before
     passing to the httpx AsyncClient.
-    For a full list of the params, see [this link](https://www.python-httpx.org/api/#asyncclient)
+    `Full list of params. <https://www.python-httpx.org/api/#asyncclient>`__
     """
 
     sync_client_kwargs: Optional[dict] = {}
     """Additional kwargs to merge with client_kwargs before
     passing to the httpx Client.
-    For a full list of the params, see [this link](https://www.python-httpx.org/api/#client)
+    `Full list of params. <https://www.python-httpx.org/api/#client>`__
     """
 
-    _client: Client = PrivateAttr(default=None)  # type: ignore
+    _client: Client = PrivateAttr()
     """
     The client to use for making requests.
     """
 
-    _async_client: AsyncClient = PrivateAttr(default=None)  # type: ignore
+    _async_client: AsyncClient = PrivateAttr()
     """
     The async client to use for making requests.
     """
@@ -473,8 +542,9 @@ class ChatOllama(BaseChatModel):
         ollama_messages = self._convert_messages_to_ollama_messages(messages)
 
         if self.stop is not None and stop is not None:
-            raise ValueError("`stop` found in both the input and default params.")
-        elif self.stop is not None:
+            msg = "`stop` found in both the input and default params."
+            raise ValueError(msg)
+        if self.stop is not None:
             stop = self.stop
 
         options_dict = kwargs.pop(
@@ -502,6 +572,7 @@ class ChatOllama(BaseChatModel):
             "messages": ollama_messages,
             "stream": kwargs.pop("stream", True),
             "model": kwargs.pop("model", self.model),
+            "think": kwargs.pop("reasoning", self.reasoning),
             "format": kwargs.pop("format", self.format),
             "options": Options(**options_dict),
             "keep_alive": kwargs.pop("keep_alive", self.keep_alive),
@@ -528,6 +599,8 @@ class ChatOllama(BaseChatModel):
 
         self._client = Client(host=self.base_url, **sync_client_kwargs)
         self._async_client = AsyncClient(host=self.base_url, **async_client_kwargs)
+        if self.validate_model_on_init:
+            validate_model(self._client, self.model)
         return self
 
     def _convert_messages_to_ollama_messages(
@@ -558,7 +631,8 @@ class ChatOllama(BaseChatModel):
                 role = "tool"
                 tool_call_id = message.tool_call_id
             else:
-                raise ValueError("Received unsupported message type for Ollama.")
+                msg = "Received unsupported message type for Ollama."
+                raise ValueError(msg)
 
             content = ""
             images = []
@@ -582,10 +656,11 @@ class ChatOllama(BaseChatModel):
                         ):
                             image_url = temp_image_url["url"]
                         else:
-                            raise ValueError(
+                            msg = (
                                 "Only string image_url or dict with string 'url' "
                                 "inside content parts are supported."
                             )
+                            raise ValueError(msg)
 
                         image_url_components = image_url.split(",")
                         # Support data:image/jpeg;base64,<image> format
@@ -598,47 +673,27 @@ class ChatOllama(BaseChatModel):
                         image = _get_image_from_data_content_block(content_part)
                         images.append(image)
                     else:
-                        raise ValueError(
+                        msg = (
                             "Unsupported message content type. "
                             "Must either have type 'text' or type 'image_url' "
                             "with a string 'image_url' field."
                         )
-            # Should convert to ollama.Message once role includes tool, and tool_call_id is in Message # noqa: E501
-            msg: dict = {
+                        raise ValueError(msg)
+            # Should convert to ollama.Message once role includes tool,
+            # and tool_call_id is in Message
+            msg_: dict = {
                 "role": role,
                 "content": content,
                 "images": images,
             }
             if tool_calls:
-                msg["tool_calls"] = tool_calls  # type: ignore
+                msg_["tool_calls"] = tool_calls
             if tool_call_id:
-                msg["tool_call_id"] = tool_call_id
-            ollama_messages.append(msg)
+                msg_["tool_call_id"] = tool_call_id
+            ollama_messages.append(msg_)
 
         return ollama_messages
 
-    def _extract_reasoning(
-        self, message_chunk: BaseMessageChunk, is_thinking: bool
-    ) -> tuple[BaseMessageChunk, bool]:
-        """Mutate a message chunk to extract reasoning content."""
-        if not self.extract_reasoning:
-            return message_chunk, is_thinking
-        elif self.extract_reasoning is True:
-            start_token = DEFAULT_THINK_TOKEN_START
-            end_token = DEFAULT_THINK_TOKEN_END
-        else:
-            start_token, end_token = cast(tuple, self.extract_reasoning)
-        if start_token in message_chunk.content:
-            is_thinking = True
-        content = message_chunk.content
-        if is_thinking:
-            message_chunk.additional_kwargs["reasoning_content"] = content
-            message_chunk.content = ""
-        if end_token in content:
-            is_thinking = False
-
-        return message_chunk, is_thinking
-
     async def _acreate_chat_stream(
         self,
         messages: list[BaseMessage],
@@ -662,16 +717,18 @@ class ChatOllama(BaseChatModel):
         chat_params = self._chat_params(messages, stop, **kwargs)
 
         if chat_params["stream"]:
-            yield from self._client.chat(**chat_params)
+            if self._client:
+                yield from self._client.chat(**chat_params)
         else:
-            yield self._client.chat(**chat_params)
+            if self._client:
+                yield self._client.chat(**chat_params)
 
     def _chat_stream_with_aggregation(
         self,
         messages: list[BaseMessage],
         stop: Optional[list[str]] = None,
         run_manager: Optional[CallbackManagerForLLMRun] = None,
-        verbose: bool = False,
+        verbose: bool = False,  # noqa: FBT001, FBT002
         **kwargs: Any,
     ) -> ChatGenerationChunk:
         final_chunk = None
@@ -687,7 +744,8 @@ class ChatOllama(BaseChatModel):
                     verbose=verbose,
                 )
         if final_chunk is None:
-            raise ValueError("No data received from Ollama stream.")
+            msg = "No data received from Ollama stream."
+            raise ValueError(msg)
 
         return final_chunk
 
@@ -696,7 +754,7 @@ class ChatOllama(BaseChatModel):
         messages: list[BaseMessage],
         stop: Optional[list[str]] = None,
         run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
-        verbose: bool = False,
+        verbose: bool = False,  # noqa: FBT001, FBT002
         **kwargs: Any,
     ) -> ChatGenerationChunk:
         final_chunk = None
@@ -712,7 +770,8 @@ class ChatOllama(BaseChatModel):
                     verbose=verbose,
                 )
         if final_chunk is None:
-            raise ValueError("No data received from Ollama stream.")
+            msg = "No data received from Ollama stream."
+            raise ValueError(msg)
 
         return final_chunk
 
@@ -759,22 +818,35 @@ class ChatOllama(BaseChatModel):
         stop: Optional[list[str]] = None,
         **kwargs: Any,
     ) -> Iterator[ChatGenerationChunk]:
-        is_thinking = False
+        reasoning = kwargs.get("reasoning", self.reasoning)
         for stream_resp in self._create_chat_stream(messages, stop, **kwargs):
             if not isinstance(stream_resp, str):
                 if stream_resp.get("done") is True:
                     generation_info = dict(stream_resp)
+                    if "model" in generation_info:
+                        generation_info["model_name"] = generation_info["model"]
                     _ = generation_info.pop("message", None)
                 else:
                     generation_info = None
+
+                content = (
+                    stream_resp["message"]["content"]
+                    if "message" in stream_resp and "content" in stream_resp["message"]
+                    else ""
+                )
+
+                additional_kwargs = {}
+                if (
+                    reasoning
+                    and "message" in stream_resp
+                    and (thinking_content := stream_resp["message"].get("thinking"))
+                ):
+                    additional_kwargs["reasoning_content"] = thinking_content
+
                 chunk = ChatGenerationChunk(
                     message=AIMessageChunk(
-                        content=(
-                            stream_resp["message"]["content"]
-                            if "message" in stream_resp
-                            and "content" in stream_resp["message"]
-                            else ""
-                        ),
+                        content=content,
+                        additional_kwargs=additional_kwargs,
                         usage_metadata=_get_usage_metadata_from_generation_info(
                             stream_resp
                         ),
@@ -782,15 +854,7 @@ class ChatOllama(BaseChatModel):
                     ),
                     generation_info=generation_info,
                 )
-                if chunk.generation_info and (
-                    model := chunk.generation_info.get("model")
-                ):
-                    chunk.generation_info["model_name"] = model  # backwards compat
-                if self.extract_reasoning:
-                    message, is_thinking = self._extract_reasoning(
-                        chunk.message, is_thinking
-                    )
-                    chunk.message = message
+
                 yield chunk
 
     def _stream(
@@ -814,22 +878,35 @@ class ChatOllama(BaseChatModel):
         stop: Optional[list[str]] = None,
         **kwargs: Any,
     ) -> AsyncIterator[ChatGenerationChunk]:
-        is_thinking = False
+        reasoning = kwargs.get("reasoning", self.reasoning)
         async for stream_resp in self._acreate_chat_stream(messages, stop, **kwargs):
             if not isinstance(stream_resp, str):
                 if stream_resp.get("done") is True:
                     generation_info = dict(stream_resp)
+                    if "model" in generation_info:
+                        generation_info["model_name"] = generation_info["model"]
                     _ = generation_info.pop("message", None)
                 else:
                     generation_info = None
+
+                content = (
+                    stream_resp["message"]["content"]
+                    if "message" in stream_resp and "content" in stream_resp["message"]
+                    else ""
+                )
+
+                additional_kwargs = {}
+                if (
+                    reasoning
+                    and "message" in stream_resp
+                    and (thinking_content := stream_resp["message"].get("thinking"))
+                ):
+                    additional_kwargs["reasoning_content"] = thinking_content
+
                 chunk = ChatGenerationChunk(
                     message=AIMessageChunk(
-                        content=(
-                            stream_resp["message"]["content"]
-                            if "message" in stream_resp
-                            and "content" in stream_resp["message"]
-                            else ""
-                        ),
+                        content=content,
+                        additional_kwargs=additional_kwargs,
                         usage_metadata=_get_usage_metadata_from_generation_info(
                             stream_resp
                         ),
@@ -837,15 +914,7 @@ class ChatOllama(BaseChatModel):
                     ),
                     generation_info=generation_info,
                 )
-                if chunk.generation_info and (
-                    model := chunk.generation_info.get("model")
-                ):
-                    chunk.generation_info["model_name"] = model  # backwards compat
-                if self.extract_reasoning:
-                    message, is_thinking = self._extract_reasoning(
-                        chunk.message, is_thinking
-                    )
-                    chunk.message = message
+
                 yield chunk
 
     async def _astream(
@@ -894,7 +963,7 @@ class ChatOllama(BaseChatModel):
         self,
         tools: Sequence[Union[dict[str, Any], type, Callable, BaseTool]],
         *,
-        tool_choice: Optional[Union[dict, str, Literal["auto", "any"], bool]] = None,
+        tool_choice: Optional[Union[dict, str, Literal["auto", "any"], bool]] = None,  # noqa: PYI051
         **kwargs: Any,
     ) -> Runnable[LanguageModelInput, BaseMessage]:
         """Bind tool-like objects to this chat model.
@@ -909,7 +978,7 @@ class ChatOllama(BaseChatModel):
                 is currently ignored as it is not supported by Ollama.**
             kwargs: Any additional parameters are passed directly to
                 ``self.bind(**kwargs)``.
-        """  # noqa: E501
+        """
         formatted_tools = [convert_to_openai_tool(tool) for tool in tools]
         return super().bind(tools=formatted_tools, **kwargs)
 
@@ -942,7 +1011,7 @@ class ChatOllama(BaseChatModel):
             method: The method for steering model generation, one of:
 
                 - "json_schema":
-                    Uses Ollama's structured output API: https://ollama.com/blog/structured-outputs
+                    Uses Ollama's `structured output API <https://ollama.com/blog/structured-outputs>`__
                 - "function_calling":
                     Uses Ollama's tool-calling API
                 - "json_mode":
@@ -1166,14 +1235,16 @@ class ChatOllama(BaseChatModel):
         """  # noqa: E501, D301
         _ = kwargs.pop("strict", None)
         if kwargs:
-            raise ValueError(f"Received unsupported arguments {kwargs}")
+            msg = f"Received unsupported arguments {kwargs}"
+            raise ValueError(msg)
         is_pydantic_schema = _is_pydantic_class(schema)
         if method == "function_calling":
             if schema is None:
-                raise ValueError(
+                msg = (
                     "schema must be specified when method is not 'json_mode'. "
                     "Received None."
                 )
+                raise ValueError(msg)
             formatted_tool = convert_to_openai_tool(schema)
             tool_name = formatted_tool["function"]["name"]
             llm = self.bind_tools(
@@ -1208,10 +1279,11 @@ class ChatOllama(BaseChatModel):
             )
         elif method == "json_schema":
             if schema is None:
-                raise ValueError(
+                msg = (
                     "schema must be specified when method is not 'json_mode'. "
                     "Received None."
                 )
+                raise ValueError(msg)
             if is_pydantic_schema:
                 schema = cast(TypeBaseModel, schema)
                 if issubclass(schema, BaseModelV1):
@@ -1225,7 +1297,7 @@ class ChatOllama(BaseChatModel):
                         "schema": schema,
                     },
                 )
-                output_parser = PydanticOutputParser(pydantic_object=schema)
+                output_parser = PydanticOutputParser(pydantic_object=schema)  # type: ignore[arg-type]
             else:
                 if is_typeddict(schema):
                     response_format = convert_to_json_schema(schema)
@@ -1245,10 +1317,11 @@ class ChatOllama(BaseChatModel):
                 )
                 output_parser = JsonOutputParser()
         else:
-            raise ValueError(
+            msg = (
                 f"Unrecognized method argument. Expected one of 'function_calling', "
                 f"'json_schema', or 'json_mode'. Received: '{method}'"
             )
+            raise ValueError(msg)
 
         if include_raw:
             parser_assign = RunnablePassthrough.assign(
@@ -1259,5 +1332,4 @@ class ChatOllama(BaseChatModel):
                 [parser_none], exception_key="parsing_error"
             )
             return RunnableMap(raw=llm) | parser_with_fallback
-        else:
-            return llm | output_parser
+        return llm | output_parser

langchain_ollama/embeddings.py

@@ -1,5 +1,7 @@
 """Ollama embeddings models."""
 
+from __future__ import annotations
+
 from typing import Any, Optional
 
 from langchain_core.embeddings import Embeddings
@@ -12,6 +14,8 @@ from pydantic import (
 )
 from typing_extensions import Self
 
+from ._utils import validate_model
+
 
 class OllamaEmbeddings(BaseModel, Embeddings):
     """Ollama embedding model integration.
@@ -95,7 +99,7 @@ class OllamaEmbeddings(BaseModel, Embeddings):
     Embed multiple texts:
         .. code-block:: python
 
-             input_texts = ["Document 1...", "Document 2..."]
+            input_texts = ["Document 1...", "Document 2..."]
             vectors = embed.embed_documents(input_texts)
             print(len(vectors))
             # The first 3 coordinates for the first vector
@@ -110,7 +114,7 @@ class OllamaEmbeddings(BaseModel, Embeddings):
         .. code-block:: python
 
             vector = await embed.aembed_query(input_text)
-           print(vector[:3])
+            print(vector[:3])
 
             # multiple:
             # await embed.aembed_documents(input_texts)
@@ -123,34 +127,38 @@ class OllamaEmbeddings(BaseModel, Embeddings):
     model: str
     """Model name to use."""
 
+    validate_model_on_init: bool = False
+    """Whether to validate the model exists in ollama locally on initialization."""
+
     base_url: Optional[str] = None
     """Base url the model is hosted under."""
 
     client_kwargs: Optional[dict] = {}
-    """Additional kwargs to pass to the httpx clients. 
+    """Additional kwargs to pass to the httpx clients.
     These arguments are passed to both synchronous and async clients.
     Use sync_client_kwargs and async_client_kwargs to pass different arguments
     to synchronous and asynchronous clients.
     """
 
     async_client_kwargs: Optional[dict] = {}
-    """Additional kwargs to merge with client_kwargs before
-    passing to the httpx AsyncClient.
-    For a full list of the params, see [this link](https://www.python-httpx.org/api/#asyncclient)
+    """Additional kwargs to merge with client_kwargs before passing to the httpx
+    AsyncClient.
+
+    For a full list of the params, see the `HTTPX documentation <https://www.python-httpx.org/api/#asyncclient>`__.
     """
 
     sync_client_kwargs: Optional[dict] = {}
-    """Additional kwargs to merge with client_kwargs before
-    passing to the httpx Client.
-    For a full list of the params, see [this link](https://www.python-httpx.org/api/#client)
+    """Additional kwargs to merge with client_kwargs before passing to the HTTPX Client.
+
+    For a full list of the params, see the `HTTPX documentation <https://www.python-httpx.org/api/#client>`__.
     """
 
-    _client: Client = PrivateAttr(default=None)  # type: ignore
+    _client: Optional[Client] = PrivateAttr(default=None)
     """
     The client to use for making requests.
     """
 
-    _async_client: AsyncClient = PrivateAttr(default=None)  # type: ignore
+    _async_client: Optional[AsyncClient] = PrivateAttr(default=None)
     """
     The async client to use for making requests.
     """
@@ -258,14 +266,21 @@ class OllamaEmbeddings(BaseModel, Embeddings):
 
         self._client = Client(host=self.base_url, **sync_client_kwargs)
         self._async_client = AsyncClient(host=self.base_url, **async_client_kwargs)
+        if self.validate_model_on_init:
+            validate_model(self._client, self.model)
         return self
 
     def embed_documents(self, texts: list[str]) -> list[list[float]]:
         """Embed search docs."""
-        embedded_docs = self._client.embed(
+        if not self._client:
+            msg = (
+                "Ollama client is not initialized. "
+                "Please ensure Ollama is running and the model is loaded."
+            )
+            raise ValueError(msg)
+        return self._client.embed(
             self.model, texts, options=self._default_params, keep_alive=self.keep_alive
         )["embeddings"]
-        return embedded_docs
 
     def embed_query(self, text: str) -> list[float]:
         """Embed query text."""
@@ -273,12 +288,17 @@ class OllamaEmbeddings(BaseModel, Embeddings):
 
     async def aembed_documents(self, texts: list[str]) -> list[list[float]]:
         """Embed search docs."""
-        embedded_docs = (
+        if not self._async_client:
+            msg = (
+                "Ollama client is not initialized. "
+                "Please ensure Ollama is running and the model is loaded."
+            )
+            raise ValueError(msg)
+        return (
             await self._async_client.embed(
                 self.model, texts, keep_alive=self.keep_alive
             )
         )["embeddings"]
-        return embedded_docs
 
     async def aembed_query(self, text: str) -> list[float]:
         """Embed query text."""

langchain_ollama/llms.py

@@ -1,5 +1,7 @@
 """Ollama large language models."""
 
+from __future__ import annotations
+
 from collections.abc import AsyncIterator, Iterator, Mapping
 from typing import (
     Any,
@@ -18,6 +20,8 @@ from ollama import AsyncClient, Client, Options
 from pydantic import PrivateAttr, model_validator
 from typing_extensions import Self
 
+from ._utils import validate_model
+
 
 class OllamaLLM(BaseLLM):
     """OllamaLLM large language models.
@@ -28,12 +32,29 @@ class OllamaLLM(BaseLLM):
             from langchain_ollama import OllamaLLM
 
             model = OllamaLLM(model="llama3")
-            model.invoke("Come up with 10 names for a song about parrots")
+            print(model.invoke("Come up with 10 names for a song about parrots"))
     """
 
     model: str
     """Model name to use."""
 
+    reasoning: Optional[bool] = None
+    """Controls the reasoning/thinking mode for
+    `supported models <https://ollama.com/search?c=thinking>`__.
+
+    - ``True``: Enables reasoning mode. The model's reasoning process will be
+      captured and returned separately in the ``additional_kwargs`` of the
+      response message, under ``reasoning_content``. The main response
+      content will not include the reasoning tags.
+    - ``False``: Disables reasoning mode. The model will not perform any reasoning,
+      and the response will not include any reasoning content.
+    - ``None`` (Default): The model will use its default reasoning behavior. If
+      the model performs reasoning, the ``<think>`` and ``</think>`` tags will
+      be present directly within the main response content."""
+
+    validate_model_on_init: bool = False
+    """Whether to validate the model exists in ollama locally on initialization."""
+
     mirostat: Optional[int] = None
     """Enable Mirostat sampling for controlling perplexity.
     (default: 0, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)"""
@@ -51,7 +72,7 @@ class OllamaLLM(BaseLLM):
 
     num_ctx: Optional[int] = None
     """Sets the size of the context window used to generate the
-    next token. (Default: 2048)	"""
+    next token. (Default: 2048)"""
 
     num_gpu: Optional[int] = None
     """The number of GPUs to use. On macOS it defaults to 1 to
@@ -113,30 +134,31 @@ class OllamaLLM(BaseLLM):
     """Base url the model is hosted under."""
 
     client_kwargs: Optional[dict] = {}
-    """Additional kwargs to pass to the httpx clients. 
+    """Additional kwargs to pass to the httpx clients.
     These arguments are passed to both synchronous and async clients.
     Use sync_client_kwargs and async_client_kwargs to pass different arguments
     to synchronous and asynchronous clients.
     """
 
     async_client_kwargs: Optional[dict] = {}
-    """Additional kwargs to merge with client_kwargs before
-    passing to the httpx AsyncClient.
-    For a full list of the params, see [this link](https://www.python-httpx.org/api/#asyncclient)
+    """Additional kwargs to merge with client_kwargs before passing to the HTTPX
+    AsyncClient.
+
+    For a full list of the params, see the `HTTPX documentation <https://www.python-httpx.org/api/#asyncclient>`__.
     """
 
     sync_client_kwargs: Optional[dict] = {}
-    """Additional kwargs to merge with client_kwargs before
-    passing to the httpx Client.
-    For a full list of the params, see [this link](https://www.python-httpx.org/api/#client)
+    """Additional kwargs to merge with client_kwargs before passing to the HTTPX Client.
+
+    For a full list of the params, see the `HTTPX documentation <https://www.python-httpx.org/api/#client>`__.
     """
 
-    _client: Client = PrivateAttr(default=None)  # type: ignore
+    _client: Optional[Client] = PrivateAttr(default=None)
     """
     The client to use for making requests.
     """
 
-    _async_client: AsyncClient = PrivateAttr(default=None)  # type: ignore
+    _async_client: Optional[AsyncClient] = PrivateAttr(default=None)
     """
     The async client to use for making requests.
     """
@@ -148,8 +170,9 @@ class OllamaLLM(BaseLLM):
         **kwargs: Any,
     ) -> dict[str, Any]:
         if self.stop is not None and stop is not None:
-            raise ValueError("`stop` found in both the input and default params.")
-        elif self.stop is not None:
+            msg = "`stop` found in both the input and default params."
+            raise ValueError(msg)
+        if self.stop is not None:
             stop = self.stop
 
         options_dict = kwargs.pop(
@@ -173,18 +196,17 @@ class OllamaLLM(BaseLLM):
             },
         )
 
-        params = {
+        return {
             "prompt": prompt,
             "stream": kwargs.pop("stream", True),
             "model": kwargs.pop("model", self.model),
+            "think": kwargs.pop("reasoning", self.reasoning),
             "format": kwargs.pop("format", self.format),
             "options": Options(**options_dict),
             "keep_alive": kwargs.pop("keep_alive", self.keep_alive),
             **kwargs,
         }
 
-        return params
-
     @property
     def _llm_type(self) -> str:
         """Return type of LLM."""
@@ -214,6 +236,8 @@ class OllamaLLM(BaseLLM):
 
         self._client = Client(host=self.base_url, **sync_client_kwargs)
         self._async_client = AsyncClient(host=self.base_url, **async_client_kwargs)
+        if self.validate_model_on_init:
+            validate_model(self._client, self.model)
         return self
 
     async def _acreate_generate_stream(
@@ -222,10 +246,11 @@ class OllamaLLM(BaseLLM):
         stop: Optional[list[str]] = None,
         **kwargs: Any,
     ) -> AsyncIterator[Union[Mapping[str, Any], str]]:
-        async for part in await self._async_client.generate(
-            **self._generate_params(prompt, stop=stop, **kwargs)
-        ):  # type: ignore
-            yield part  # type: ignore
+        if self._async_client:
+            async for part in await self._async_client.generate(
+                **self._generate_params(prompt, stop=stop, **kwargs)
+            ):
+                yield part
 
     def _create_generate_stream(
         self,
@@ -233,23 +258,27 @@ class OllamaLLM(BaseLLM):
         stop: Optional[list[str]] = None,
         **kwargs: Any,
     ) -> Iterator[Union[Mapping[str, Any], str]]:
-        yield from self._client.generate(
-            **self._generate_params(prompt, stop=stop, **kwargs)
-        )  # type: ignore
+        if self._client:
+            yield from self._client.generate(
+                **self._generate_params(prompt, stop=stop, **kwargs)
+            )
 
     async def _astream_with_aggregation(
         self,
         prompt: str,
         stop: Optional[list[str]] = None,
         run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
-        verbose: bool = False,
+        verbose: bool = False,  # noqa: FBT001, FBT002
         **kwargs: Any,
     ) -> GenerationChunk:
         final_chunk = None
+        thinking_content = ""
         async for stream_resp in self._acreate_generate_stream(prompt, stop, **kwargs):
             if not isinstance(stream_resp, str):
+                if stream_resp.get("thinking"):
+                    thinking_content += stream_resp["thinking"]
                 chunk = GenerationChunk(
-                    text=stream_resp["response"] if "response" in stream_resp else "",
+                    text=stream_resp.get("response", ""),
                     generation_info=(
                         dict(stream_resp) if stream_resp.get("done") is True else None
                     ),
@@ -265,7 +294,14 @@ class OllamaLLM(BaseLLM):
                         verbose=verbose,
                     )
         if final_chunk is None:
-            raise ValueError("No data received from Ollama stream.")
+            msg = "No data received from Ollama stream."
+            raise ValueError(msg)
+
+        if thinking_content:
+            if final_chunk.generation_info:
+                final_chunk.generation_info["thinking"] = thinking_content
+            else:
+                final_chunk.generation_info = {"thinking": thinking_content}
 
         return final_chunk
 
@@ -274,14 +310,17 @@ class OllamaLLM(BaseLLM):
         prompt: str,
         stop: Optional[list[str]] = None,
         run_manager: Optional[CallbackManagerForLLMRun] = None,
-        verbose: bool = False,
+        verbose: bool = False,  # noqa: FBT001, FBT002
         **kwargs: Any,
     ) -> GenerationChunk:
         final_chunk = None
+        thinking_content = ""
         for stream_resp in self._create_generate_stream(prompt, stop, **kwargs):
             if not isinstance(stream_resp, str):
+                if stream_resp.get("thinking"):
+                    thinking_content += stream_resp["thinking"]
                 chunk = GenerationChunk(
-                    text=stream_resp["response"] if "response" in stream_resp else "",
+                    text=stream_resp.get("response", ""),
                     generation_info=(
                         dict(stream_resp) if stream_resp.get("done") is True else None
                     ),
@@ -297,7 +336,14 @@ class OllamaLLM(BaseLLM):
                         verbose=verbose,
                     )
         if final_chunk is None:
-            raise ValueError("No data received from Ollama stream.")
+            msg = "No data received from Ollama stream."
+            raise ValueError(msg)
+
+        if thinking_content:
+            if final_chunk.generation_info:
+                final_chunk.generation_info["thinking"] = thinking_content
+            else:
+                final_chunk.generation_info = {"thinking": thinking_content}
 
         return final_chunk
 
@@ -346,13 +392,22 @@ class OllamaLLM(BaseLLM):
         run_manager: Optional[CallbackManagerForLLMRun] = None,
         **kwargs: Any,
     ) -> Iterator[GenerationChunk]:
+        reasoning = kwargs.get("reasoning", self.reasoning)
         for stream_resp in self._create_generate_stream(prompt, stop, **kwargs):
             if not isinstance(stream_resp, str):
+                additional_kwargs = {}
+                if reasoning and (thinking_content := stream_resp.get("thinking")):
+                    additional_kwargs["reasoning_content"] = thinking_content
+
                 chunk = GenerationChunk(
                     text=(stream_resp.get("response", "")),
-                    generation_info=(
-                        dict(stream_resp) if stream_resp.get("done") is True else None
-                    ),
+                    generation_info={
+                        "finish_reason": self.stop,
+                        **additional_kwargs,
+                        **(
+                            dict(stream_resp) if stream_resp.get("done") is True else {}
+                        ),
+                    },
                 )
                 if run_manager:
                     run_manager.on_llm_new_token(
@@ -368,13 +423,22 @@ class OllamaLLM(BaseLLM):
         run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
         **kwargs: Any,
     ) -> AsyncIterator[GenerationChunk]:
+        reasoning = kwargs.get("reasoning", self.reasoning)
         async for stream_resp in self._acreate_generate_stream(prompt, stop, **kwargs):
             if not isinstance(stream_resp, str):
+                additional_kwargs = {}
+                if reasoning and (thinking_content := stream_resp.get("thinking")):
+                    additional_kwargs["reasoning_content"] = thinking_content
+
                 chunk = GenerationChunk(
                     text=(stream_resp.get("response", "")),
-                    generation_info=(
-                        dict(stream_resp) if stream_resp.get("done") is True else None
-                    ),
+                    generation_info={
+                        "finish_reason": self.stop,
+                        **additional_kwargs,
+                        **(
+                            dict(stream_resp) if stream_resp.get("done") is True else {}
+                        ),
+                    },
                 )
                 if run_manager:
                     await run_manager.on_llm_new_token(

{langchain_ollama-0.3.3.dist-info → langchain_ollama-0.3.4.dist-info}/METADATA

@@ -1,14 +1,14 @@
 Metadata-Version: 2.1
 Name: langchain-ollama
-Version: 0.3.3
+Version: 0.3.4
 Summary: An integration package connecting Ollama and LangChain
 License: MIT
 Project-URL: Source Code, https://github.com/langchain-ai/langchain/tree/master/libs/partners/ollama
 Project-URL: Release Notes, https://github.com/langchain-ai/langchain/releases?q=tag%3A%22langchain-ollama%3D%3D0%22&expanded=true
 Project-URL: repository, https://github.com/langchain-ai/langchain
 Requires-Python: >=3.9
-Requires-Dist: ollama<1.0.0,>=0.4.8
-Requires-Dist: langchain-core<1.0.0,>=0.3.60
+Requires-Dist: ollama<1.0.0,>=0.5.1
+Requires-Dist: langchain-core<1.0.0,>=0.3.68
 Description-Content-Type: text/markdown
 
 # langchain-ollama
@@ -21,37 +21,48 @@ This package contains the LangChain integration with Ollama
 pip install -U langchain-ollama
 ```
 
-You will also need to run the Ollama server locally. 
-You can download it [here](https://ollama.com/download).
+For the package to work, you will need to install and run the Ollama server locally ([download](https://ollama.com/download)).
 
-## Chat Models
+To run integration tests (`make integration_tests`), you will need the following models installed in your Ollama server:
+
+- `llama3.1`
+- `deepseek-r1:1.5b`
+
+Install these models by running:
+
+```bash
+ollama pull <name-of-model>
+```
+
+## [Chat Models](https://python.langchain.com/api_reference/ollama/chat_models/langchain_ollama.chat_models.ChatOllama.html#chatollama)
 
 `ChatOllama` class exposes chat models from Ollama.
 
 ```python
 from langchain_ollama import ChatOllama
 
-llm = ChatOllama(model="llama3-groq-tool-use")
+llm = ChatOllama(model="llama3.1")
 llm.invoke("Sing a ballad of LangChain.")
 ```
 
-## Embeddings
+## [Embeddings](https://python.langchain.com/api_reference/ollama/embeddings/langchain_ollama.embeddings.OllamaEmbeddings.html#ollamaembeddings)
 
 `OllamaEmbeddings` class exposes embeddings from Ollama.
 
 ```python
 from langchain_ollama import OllamaEmbeddings
 
-embeddings = OllamaEmbeddings(model="llama3")
+embeddings = OllamaEmbeddings(model="llama3.1")
 embeddings.embed_query("What is the meaning of life?")
 ```
 
-## LLMs
-`OllamaLLM` class exposes LLMs from Ollama.
+## [LLMs](https://python.langchain.com/api_reference/ollama/llms/langchain_ollama.llms.OllamaLLM.html#ollamallm)
+
+`OllamaLLM` class exposes traditional LLMs from Ollama.
 
 ```python
 from langchain_ollama import OllamaLLM
 
-llm = OllamaLLM(model="llama3")
+llm = OllamaLLM(model="llama3.1")
 llm.invoke("The meaning of life is")
 ```

langchain_ollama-0.3.4.dist-info/RECORD

@@ -0,0 +1,11 @@
+langchain_ollama-0.3.4.dist-info/METADATA,sha256=wM54qEosykpO89kExse0V4Y3K3ncspLP_mFNKsBxTNY,2072
+langchain_ollama-0.3.4.dist-info/WHEEL,sha256=9P2ygRxDrTJz3gsagc0Z96ukrxjr-LFBGOgv3AuKlCA,90
+langchain_ollama-0.3.4.dist-info/entry_points.txt,sha256=6OYgBcLyFCUgeqLgnvMyOJxPCWzgy7se4rLPKtNonMs,34
+langchain_ollama-0.3.4.dist-info/licenses/LICENSE,sha256=2btS8uNUDWD_UNjw9ba6ZJt_00aUjEw9CGyK-xIHY8c,1072
+langchain_ollama/__init__.py,sha256=TI1gI0Wpg7mRXehGpxrJG2flF_t4Ev-aIJlLKV-CgL0,633
+langchain_ollama/_utils.py,sha256=dmFO4tSvDTeMALc89QnTBLNWPMZL0eNAq1EDwuMjRA8,1416
+langchain_ollama/chat_models.py,sha256=olz3KJeLG1vk47Xl38nN9bP4bcol5cBQnPnu5MyP8k8,55539
+langchain_ollama/embeddings.py,sha256=VprOFiBRuUPGEygoIfxvAZStUsqRj65ZNMpkvCAo_9Y,10239
+langchain_ollama/llms.py,sha256=PSJ-VQMocp1nm-pgtnKnozidt66RKJiEnhdzftoLNNc,16778
+langchain_ollama/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+langchain_ollama-0.3.4.dist-info/RECORD,,

{langchain_ollama-0.3.3.dist-info → langchain_ollama-0.3.4.dist-info}/WHEEL

@@ -1,4 +1,4 @@
 Wheel-Version: 1.0
-Generator: pdm-backend (2.4.4)
+Generator: pdm-backend (2.4.5)
 Root-Is-Purelib: true
 Tag: py3-none-any

langchain_ollama-0.3.3.dist-info/RECORD

@@ -1,10 +0,0 @@
-langchain_ollama-0.3.3.dist-info/METADATA,sha256=K2QhMD3eEMIMegVdXf6ZyQ7C5fbl2wQ1CvvqtUOmyug,1462
-langchain_ollama-0.3.3.dist-info/WHEEL,sha256=tSfRZzRHthuv7vxpI4aehrdN9scLjk-dCJkPLzkHxGg,90
-langchain_ollama-0.3.3.dist-info/entry_points.txt,sha256=6OYgBcLyFCUgeqLgnvMyOJxPCWzgy7se4rLPKtNonMs,34
-langchain_ollama-0.3.3.dist-info/licenses/LICENSE,sha256=2btS8uNUDWD_UNjw9ba6ZJt_00aUjEw9CGyK-xIHY8c,1072
-langchain_ollama/__init__.py,sha256=1f8Cyf1_bS0CT16U8-Os1P1Oa3erIDtIBTH4KVmBLvY,633
-langchain_ollama/chat_models.py,sha256=Z2wzR5R568aNyH1LKN84kUdNZFOvvgY-csE626_sBVc,51723
-langchain_ollama/embeddings.py,sha256=udL26XHdUMybQogY9Gj3vlJXxxkVAVZ-9He2U8wlJ3k,9547
-langchain_ollama/llms.py,sha256=Rin6HVZvrH1epRsjhojSmOBFWAaU0cfOU1gV6I0bqJE,13933
-langchain_ollama/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-langchain_ollama-0.3.3.dist-info/RECORD,,