langchain-dev-utils 1.2.9__py3-none-any.whl → 1.2.10__py3-none-any.whl

langchain_dev_utils/__init__.py

@@ -1 +1 @@
-__version__ = "1.2.9"
+__version__ = "1.2.10"

langchain_dev_utils/chat_models/adapters/openai_compatible.py

@@ -1,572 +1,584 @@
-from __future__ import annotations
-
-from collections.abc import AsyncIterator, Iterator
-from json import JSONDecodeError
-from typing import (
-    Any,
-    Callable,
-    List,
-    Literal,
-    Optional,
-    Sequence,
-    Type,
-    TypeVar,
-    Union,
-)
-
-import openai
-from langchain_core.callbacks import (
-    AsyncCallbackManagerForLLMRun,
-    CallbackManagerForLLMRun,
-)
-from langchain_core.language_models import LangSmithParams, LanguageModelInput
-from langchain_core.messages import AIMessage, AIMessageChunk, BaseMessage, HumanMessage
-from langchain_core.outputs import ChatGenerationChunk, ChatResult
-from langchain_core.runnables import Runnable
-from langchain_core.tools import BaseTool
-from langchain_core.utils import from_env, secret_from_env
-from langchain_core.utils.function_calling import convert_to_openai_tool
-from langchain_openai.chat_models._compat import _convert_from_v1_to_chat_completions
-from langchain_openai.chat_models.base import BaseChatOpenAI, _convert_message_to_dict
-from pydantic import (
-    BaseModel,
-    ConfigDict,
-    Field,
-    PrivateAttr,
-    SecretStr,
-    create_model,
-    model_validator,
-)
-from typing_extensions import Self
-
-from ..types import (
-    CompatibilityOptions,
-    ReasoningKeepPolicy,
-    ResponseFormatType,
-    ToolChoiceType,
-)
-
-_BM = TypeVar("_BM", bound=BaseModel)
-_DictOrPydanticClass = Union[dict[str, Any], type[_BM], type]
-_DictOrPydantic = Union[dict, _BM]
-
-
-def _get_last_human_message_index(messages: list[BaseMessage]) -> int:
-    """find the index of the last HumanMessage in the messages list, return -1 if not found."""
-    return next(
-        (
-            i
-            for i in range(len(messages) - 1, -1, -1)
-            if isinstance(messages[i], HumanMessage)
-        ),
-        -1,
-    )
-
-
-class _BaseChatOpenAICompatible(BaseChatOpenAI):
-    """
-    Base template class for OpenAI-compatible chat model implementations.
-
-    This class provides a foundation for integrating various LLM providers that
-    offer OpenAI-compatible APIs (such as vLLM, OpenRouter, ZAI, Moonshot,
-    and many others). It enhances the base OpenAI functionality by:
-
-    **1. Supports output of more types of reasoning content (reasoning_content)**
-    ChatOpenAI can only output reasoning content natively supported by official
-    OpenAI models, while OpenAICompatibleChatModel can output reasoning content
-    from other model providers (e.g., OpenRouter, vLLM).
-
-    **2. Dynamically adapts to choose the most suitable structured-output method**
-    OpenAICompatibleChatModel adds method="auto" (default), which selects the best
-    structured-output method (function_calling or json_schema) based on the actual
-    capabilities of the model provider.
-
-    **3. Supports configuration of related parameters**
-    For cases where parameters differ from the official OpenAI API, this library
-    provides the compatibility_options parameter to address this issue. For
-    example, when different model providers have inconsistent support for
-    tool_choice, you can adapt by setting supported_tool_choice in
-    compatibility_options.
-
-    Built on top of `langchain-openai`'s `BaseChatOpenAI`, this template class
-    extends capabilities to better support diverse OpenAI-compatible model
-    providers while maintaining full compatibility with LangChain's chat model
-    interface.
-
-    Note: This is a template class and should not be exported or instantiated
-    directly. Instead, use it as a base class and provide the specific provider
-    name through inheritance or the factory function
-    `_create_openai_compatible_model()`.
-    """
-
-    model_name: str = Field(alias="model", default="openai compatible model")
-    """The name of the model"""
-    api_key: Optional[SecretStr] = Field(
-        default_factory=secret_from_env("OPENAI_COMPATIBLE_API_KEY", default=None),
-    )
-    """OpenAI Compatible API key"""
-    api_base: str = Field(
-        default_factory=from_env("OPENAI_COMPATIBLE_API_BASE", default=""),
-    )
-    """OpenAI Compatible API base URL"""
-
-    model_config = ConfigDict(populate_by_name=True)
-
-    _provider: str = PrivateAttr(default="openai-compatible")
-
-    """Provider Compatibility Options"""
-    supported_tool_choice: ToolChoiceType = Field(default_factory=list)
-    """Supported tool choice"""
-    supported_response_format: ResponseFormatType = Field(default_factory=list)
-    """Supported response format"""
-    reasoning_keep_policy: ReasoningKeepPolicy = Field(default="never")
-    """How to keep reasoning content in the messages"""
-    include_usage: bool = Field(default=True)
-    """Whether to include usage information in the output"""
-
-    @property
-    def _llm_type(self) -> str:
-        return f"chat-{self._provider}"
-
-    @property
-    def lc_secrets(self) -> dict[str, str]:
-        return {"api_key": f"{self._provider.upper()}_API_KEY"}
-
-    def _get_ls_params(
-        self,
-        stop: Optional[list[str]] = None,
-        **kwargs: Any,
-    ) -> LangSmithParams:
-        ls_params = super()._get_ls_params(stop=stop, **kwargs)
-        ls_params["ls_provider"] = self._provider
-        return ls_params
-
-    def _get_request_payload(
-        self,
-        input_: LanguageModelInput,
-        *,
-        stop: list[str] | None = None,
-        **kwargs: Any,
-    ) -> dict:
-        payload = {**self._default_params, **kwargs}
-
-        if self._use_responses_api(payload):
-            return super()._get_request_payload(input_, stop=stop, **kwargs)
-
-        messages = self._convert_input(input_).to_messages()
-        if stop is not None:
-            kwargs["stop"] = stop
-
-        payload_messages = []
-        last_human_index = -1
-        if self.reasoning_keep_policy == "current":
-            last_human_index = _get_last_human_message_index(messages)
-
-        for index, m in enumerate(messages):
-            if isinstance(m, AIMessage):
-                msg_dict = _convert_message_to_dict(
-                    _convert_from_v1_to_chat_completions(m)
-                )
-                if self.reasoning_keep_policy == "all" and m.additional_kwargs.get(
-                    "reasoning_content"
-                ):
-                    msg_dict["reasoning_content"] = m.additional_kwargs.get(
-                        "reasoning_content"
-                    )
-                elif (
-                    self.reasoning_keep_policy == "current"
-                    and index > last_human_index
-                    and m.additional_kwargs.get("reasoning_content")
-                ):
-                    msg_dict["reasoning_content"] = m.additional_kwargs.get(
-                        "reasoning_content"
-                    )
-                payload_messages.append(msg_dict)
-            else:
-                payload_messages.append(_convert_message_to_dict(m))
-
-        payload["messages"] = payload_messages
-        return payload
-
-    @model_validator(mode="after")
-    def validate_environment(self) -> Self:
-        if not (self.api_key and self.api_key.get_secret_value()):
-            msg = f"{self._provider.upper()}_API_KEY must be set."
-            raise ValueError(msg)
-        client_params: dict = {
-            k: v
-            for k, v in {
-                "api_key": self.api_key.get_secret_value() if self.api_key else None,
-                "base_url": self.api_base,
-                "timeout": self.request_timeout,
-                "max_retries": self.max_retries,
-                "default_headers": self.default_headers,
-                "default_query": self.default_query,
-            }.items()
-            if v is not None
-        }
-
-        if not (self.client or None):
-            sync_specific: dict = {"http_client": self.http_client}
-            self.root_client = openai.OpenAI(**client_params, **sync_specific)
-            self.client = self.root_client.chat.completions
-        if not (self.async_client or None):
-            async_specific: dict = {"http_client": self.http_async_client}
-            self.root_async_client = openai.AsyncOpenAI(
-                **client_params,
-                **async_specific,
-            )
-            self.async_client = self.root_async_client.chat.completions
-        return self
-
-    @model_validator(mode="after")
-    def _set_model_profile(self) -> Self:
-        """Set model profile if not overridden."""
-        if self.profile is None:
-            self.profile = {}
-        return self
-
-    def _create_chat_result(
-        self,
-        response: Union[dict, openai.BaseModel],
-        generation_info: Optional[dict] = None,
-    ) -> ChatResult:
-        """Convert API response to LangChain ChatResult with enhanced content processing.
-
-        Extends base implementation to capture and preserve reasoning content from
-        model responses, supporting advanced models that provide reasoning chains
-        or thought processes alongside regular responses.
-
-        Handles multiple response formats:
-        - Standard OpenAI response objects with `reasoning_content` attribute
-        - Responses with `model_extra` containing reasoning data
-        - Dictionary responses (pass-through to base implementation)
-
-        Args:
-            response: Raw API response (OpenAI object or dict)
-            generation_info: Additional generation metadata
-
-        Returns:
-            ChatResult with enhanced message containing reasoning content when available
-        """
-        rtn = super()._create_chat_result(response, generation_info)
-
-        if not isinstance(response, openai.BaseModel):
-            return rtn
-
-        for generation in rtn.generations:
-            if generation.message.response_metadata is None:
-                generation.message.response_metadata = {}
-            generation.message.response_metadata["model_provider"] = "openai-compatible"
-
-        choices = getattr(response, "choices", None)
-        if choices and hasattr(choices[0].message, "reasoning_content"):
-            rtn.generations[0].message.additional_kwargs["reasoning_content"] = choices[
-                0
-            ].message.reasoning_content
-        elif choices and hasattr(choices[0].message, "model_extra"):
-            model_extra = choices[0].message.model_extra
-            if isinstance(model_extra, dict) and (
-                reasoning := model_extra.get("reasoning")
-            ):
-                rtn.generations[0].message.additional_kwargs["reasoning_content"] = (
-                    reasoning
-                )
-
-        return rtn
-
-    def _convert_chunk_to_generation_chunk(
-        self,
-        chunk: dict,
-        default_chunk_class: type,
-        base_generation_info: Optional[dict],
-    ) -> Optional[ChatGenerationChunk]:
-        """Convert streaming chunk to generation chunk with reasoning content support.
-
-        Processes streaming response chunks to extract reasoning content alongside
-        regular message content, enabling real-time streaming of both response
-        text and reasoning chains from compatible models.
-
-        Args:
-            chunk: Raw streaming chunk from API
-            default_chunk_class: Expected chunk type for validation
-            base_generation_info: Base metadata for the generation
-
-        Returns:
-            ChatGenerationChunk with reasoning content when present in chunk data
-        """
-        generation_chunk = super()._convert_chunk_to_generation_chunk(
-            chunk,
-            default_chunk_class,
-            base_generation_info,
-        )
-        if (choices := chunk.get("choices")) and generation_chunk:
-            top = choices[0]
-            if isinstance(generation_chunk.message, AIMessageChunk):
-                generation_chunk.message.response_metadata = {
-                    **generation_chunk.message.response_metadata,
-                    "model_provider": "openai-compatible",
-                }
-                if (
-                    reasoning_content := top.get("delta", {}).get("reasoning_content")
-                ) is not None:
-                    generation_chunk.message.additional_kwargs["reasoning_content"] = (
-                        reasoning_content
-                    )
-                elif (reasoning := top.get("delta", {}).get("reasoning")) is not None:
-                    generation_chunk.message.additional_kwargs["reasoning_content"] = (
-                        reasoning
-                    )
-
-        return generation_chunk
-
-    def _stream(
-        self,
-        messages: List[BaseMessage],
-        stop: Optional[List[str]] = None,
-        run_manager: Optional[CallbackManagerForLLMRun] = None,
-        **kwargs: Any,
-    ) -> Iterator[ChatGenerationChunk]:
-        if self.include_usage:
-            kwargs["stream_options"] = {"include_usage": True}
-        try:
-            for chunk in super()._stream(
-                messages, stop=stop, run_manager=run_manager, **kwargs
-            ):
-                yield chunk
-        except JSONDecodeError as e:
-            raise JSONDecodeError(
-                f"{self._provider.title()} API returned an invalid response. "
-                "Please check the API status and try again.",
-                e.doc,
-                e.pos,
-            ) from e
-
-    async def _astream(
-        self,
-        messages: List[BaseMessage],
-        stop: Optional[List[str]] = None,
-        run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
-        **kwargs: Any,
-    ) -> AsyncIterator[ChatGenerationChunk]:
-        if self.include_usage:
-            kwargs["stream_options"] = {"include_usage": True}
-        try:
-            async for chunk in super()._astream(
-                messages, stop=stop, run_manager=run_manager, **kwargs
-            ):
-                yield chunk
-        except JSONDecodeError as e:
-            raise JSONDecodeError(
-                f"{self._provider.title()} API returned an invalid response. "
-                "Please check the API status and try again.",
-                e.doc,
-                e.pos,
-            ) from e
-
-    def _generate(
-        self,
-        messages: List[BaseMessage],
-        stop: Optional[List[str]] = None,
-        run_manager: Optional[CallbackManagerForLLMRun] = None,
-        **kwargs: Any,
-    ) -> ChatResult:
-        try:
-            return super()._generate(
-                messages, stop=stop, run_manager=run_manager, **kwargs
-            )
-        except JSONDecodeError as e:
-            raise JSONDecodeError(
-                f"{self._provider.title()} API returned an invalid response. "
-                "Please check the API status and try again.",
-                e.doc,
-                e.pos,
-            ) from e
-
-    async def _agenerate(
-        self,
-        messages: List[BaseMessage],
-        stop: Optional[List[str]] = None,
-        run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
-        **kwargs: Any,
-    ) -> ChatResult:
-        try:
-            return await super()._agenerate(
-                messages, stop=stop, run_manager=run_manager, **kwargs
-            )
-        except JSONDecodeError as e:
-            raise JSONDecodeError(
-                f"{self._provider.title()} API returned an invalid response. "
-                "Please check the API status and try again.",
-                e.doc,
-                e.pos,
-            ) from e
-
-    def bind_tools(
-        self,
-        tools: Sequence[dict[str, Any] | type | Callable | BaseTool],
-        *,
-        tool_choice: dict | str | bool | None = None,
-        strict: bool | None = None,
-        parallel_tool_calls: bool | None = None,
-        **kwargs: Any,
-    ) -> Runnable[LanguageModelInput, AIMessage]:
-        if parallel_tool_calls is not None:
-            kwargs["parallel_tool_calls"] = parallel_tool_calls
-        formatted_tools = [
-            convert_to_openai_tool(tool, strict=strict) for tool in tools
-        ]
-
-        tool_names = []
-        for tool in formatted_tools:
-            if "function" in tool:
-                tool_names.append(tool["function"]["name"])
-            elif "name" in tool:
-                tool_names.append(tool["name"])
-            else:
-                pass
-
-        support_tool_choice = False
-        if tool_choice is not None:
-            if isinstance(tool_choice, bool):
-                tool_choice = "required"
-            if isinstance(tool_choice, str):
-                if (
-                    tool_choice in ["auto", "none", "required"]
-                    and tool_choice in self.supported_tool_choice
-                ):
-                    support_tool_choice = True
-
-                elif "specific" in self.supported_tool_choice:
-                    if tool_choice in tool_names:
-                        support_tool_choice = True
-                        tool_choice = {
-                            "type": "function",
-                            "function": {"name": tool_choice},
-                        }
-            tool_choice = tool_choice if support_tool_choice else None
-        if tool_choice:
-            kwargs["tool_choice"] = tool_choice
-        return super().bind(tools=formatted_tools, **kwargs)
-
-    def with_structured_output(
-        self,
-        schema: Optional[_DictOrPydanticClass] = None,
-        *,
-        method: Literal[
-            "auto",
-            "function_calling",
-            "json_mode",
-            "json_schema",
-        ] = "auto",
-        include_raw: bool = False,
-        strict: Optional[bool] = None,
-        **kwargs: Any,
-    ) -> Runnable[LanguageModelInput, _DictOrPydantic]:
-        """Configure structured output extraction with provider compatibility handling.
-
-        Enables parsing of model outputs into structured formats (Pydantic models
-        or dictionaries) while handling provider-specific method compatibility.
-        Falls back from json_schema to function_calling for providers that don't
-        support the json_schema method.
-
-        Args:
-            schema: Output schema (Pydantic model class or dictionary definition)
-            method: Extraction method - defaults to auto,it will choice best method based on provider supported response format
-            include_raw: Whether to include raw model response alongside parsed output
-            strict: Schema enforcement strictness (provider-dependent)
-            **kwargs: Additional structured output parameters
-
-        Returns:
-            Runnable configured for structured output extraction
-        """
-        if method not in ["auto", "function_calling", "json_mode", "json_schema"]:
-            raise ValueError(
-                f"Unsupported method: {method}. Please choose from 'auto', 'function_calling', 'json_mode', 'json_schema'."
-            )
-        if method == "auto":
-            if "json_schema" in self.supported_response_format:
-                method = "json_schema"
-            else:
-                method = "function_calling"
-        elif (
-            method == "json_schema"
-            and "json_schema" not in self.supported_response_format
-        ):
-            method = "function_calling"
-        elif (
-            method == "json_mode" and "json_mode" not in self.supported_response_format
-        ):
-            method = "function_calling"
-
-        return super().with_structured_output(
-            schema,
-            method=method,
-            include_raw=include_raw,
-            strict=strict,
-            **kwargs,
-        )
-
-
-def _create_openai_compatible_model(
-    provider: str,
-    base_url: str,
-    compatibility_options: Optional[CompatibilityOptions] = None,
-) -> Type[_BaseChatOpenAICompatible]:
-    """Factory function for creating provider-specific OpenAI-compatible model classes.
-
-    Dynamically generates model classes for different OpenAI-compatible providers,
-    configuring environment variable mappings and default base URLs specific to each provider.
-
-    Args:
-        provider: Provider identifier (e.g., `vllm`,`openrouter`)
-        base_url: Default API base URL for the provider
-        compatibility_options: Optional configuration for the provider
-
-    Returns:
-        Configured model class ready for instantiation with provider-specific settings
-    """
-    chat_model_cls_name = f"Chat{provider.title()}"
-    if compatibility_options is None:
-        compatibility_options = {}
-
-    return create_model(
-        chat_model_cls_name,
-        __base__=_BaseChatOpenAICompatible,
-        api_base=(
-            str,
-            Field(
-                default_factory=from_env(
-                    f"{provider.upper()}_API_BASE", default=base_url
-                ),
-            ),
-        ),
-        api_key=(
-            str,
-            Field(
-                default_factory=secret_from_env(
-                    f"{provider.upper()}_API_KEY", default=None
-                ),
-            ),
-        ),
-        _provider=(
-            str,
-            PrivateAttr(default=provider),
-        ),
-        supported_tool_choice=(
-            ToolChoiceType,
-            Field(default=compatibility_options.get("supported_tool_choice", ["auto"])),
-        ),
-        reasoning_keep_policy=(
-            ReasoningKeepPolicy,
-            Field(default=compatibility_options.get("reasoning_keep_policy", "never")),
-        ),
-        supported_response_format=(
-            ResponseFormatType,
-            Field(default=compatibility_options.get("supported_response_format", [])),
-        ),
-        include_usage=(
-            bool,
-            Field(default=compatibility_options.get("include_usage", True)),
-        ),
-    )
+from __future__ import annotations
+
+from collections.abc import AsyncIterator, Iterator
+from json import JSONDecodeError
+from typing import (
+    Any,
+    Callable,
+    List,
+    Literal,
+    Optional,
+    Sequence,
+    Type,
+    TypeVar,
+    Union,
+)
+
+import openai
+from langchain_core.callbacks import (
+    AsyncCallbackManagerForLLMRun,
+    CallbackManagerForLLMRun,
+)
+from langchain_core.language_models import LangSmithParams, LanguageModelInput
+from langchain_core.messages import AIMessage, AIMessageChunk, BaseMessage, HumanMessage
+from langchain_core.outputs import ChatGenerationChunk, ChatResult
+from langchain_core.runnables import Runnable
+from langchain_core.tools import BaseTool
+from langchain_core.utils import from_env, secret_from_env
+from langchain_core.utils.function_calling import convert_to_openai_tool
+from langchain_openai.chat_models._compat import _convert_from_v1_to_chat_completions
+from langchain_openai.chat_models.base import BaseChatOpenAI, _convert_message_to_dict
+from pydantic import (
+    BaseModel,
+    ConfigDict,
+    Field,
+    PrivateAttr,
+    SecretStr,
+    create_model,
+    model_validator,
+)
+from typing_extensions import Self
+
+from ..types import (
+    CompatibilityOptions,
+    ReasoningKeepPolicy,
+    ResponseFormatType,
+    ToolChoiceType,
+)
+
+_BM = TypeVar("_BM", bound=BaseModel)
+_DictOrPydanticClass = Union[dict[str, Any], type[_BM], type]
+_DictOrPydantic = Union[dict, _BM]
+
+
+def _get_last_human_message_index(messages: list[BaseMessage]) -> int:
+    """find the index of the last HumanMessage in the messages list, return -1 if not found."""
+    return next(
+        (
+            i
+            for i in range(len(messages) - 1, -1, -1)
+            if isinstance(messages[i], HumanMessage)
+        ),
+        -1,
+    )
+
+
+class _BaseChatOpenAICompatible(BaseChatOpenAI):
+    """
+    Base template class for OpenAI-compatible chat model implementations.
+
+    This class provides a foundation for integrating various LLM providers that
+    offer OpenAI-compatible APIs (such as vLLM, OpenRouter, ZAI, Moonshot,
+    and many others). It enhances the base OpenAI functionality by:
+
+    **1. Supports output of more types of reasoning content (reasoning_content)**
+    ChatOpenAI can only output reasoning content natively supported by official
+    OpenAI models, while OpenAICompatibleChatModel can output reasoning content
+    from other model providers (e.g., OpenRouter, vLLM).
+
+    **2. Dynamically adapts to choose the most suitable structured-output method**
+    OpenAICompatibleChatModel adds method="auto" (default), which selects the best
+    structured-output method (function_calling or json_schema) based on the actual
+    capabilities of the model provider.
+
+    **3. Supports configuration of related parameters**
+    For cases where parameters differ from the official OpenAI API, this library
+    provides the compatibility_options parameter to address this issue. For
+    example, when different model providers have inconsistent support for
+    tool_choice, you can adapt by setting supported_tool_choice in
+    compatibility_options.
+
+    Built on top of `langchain-openai`'s `BaseChatOpenAI`, this template class
+    extends capabilities to better support diverse OpenAI-compatible model
+    providers while maintaining full compatibility with LangChain's chat model
+    interface.
+
+    Note: This is a template class and should not be exported or instantiated
+    directly. Instead, use it as a base class and provide the specific provider
+    name through inheritance or the factory function
+    `_create_openai_compatible_model()`.
+    """
+
+    model_name: str = Field(alias="model", default="openai compatible model")
+    """The name of the model"""
+    api_key: Optional[SecretStr] = Field(
+        default_factory=secret_from_env("OPENAI_COMPATIBLE_API_KEY", default=None),
+    )
+    """OpenAI Compatible API key"""
+    api_base: str = Field(
+        default_factory=from_env("OPENAI_COMPATIBLE_API_BASE", default=""),
+    )
+    """OpenAI Compatible API base URL"""
+
+    model_config = ConfigDict(populate_by_name=True)
+
+    _provider: str = PrivateAttr(default="openai-compatible")
+
+    """Provider Compatibility Options"""
+    supported_tool_choice: ToolChoiceType = Field(default_factory=list)
+    """Supported tool choice"""
+    supported_response_format: ResponseFormatType = Field(default_factory=list)
+    """Supported response format"""
+    reasoning_keep_policy: ReasoningKeepPolicy = Field(default="never")
+    """How to keep reasoning content in the messages"""
+    include_usage: bool = Field(default=True)
+    """Whether to include usage information in the output"""
+
+    @property
+    def _llm_type(self) -> str:
+        return f"chat-{self._provider}"
+
+    @property
+    def lc_secrets(self) -> dict[str, str]:
+        return {"api_key": f"{self._provider.upper()}_API_KEY"}
+
+    def _get_ls_params(
+        self,
+        stop: Optional[list[str]] = None,
+        **kwargs: Any,
+    ) -> LangSmithParams:
+        ls_params = super()._get_ls_params(stop=stop, **kwargs)
+        ls_params["ls_provider"] = self._provider
+        return ls_params
+
+    def _get_request_payload(
+        self,
+        input_: LanguageModelInput,
+        *,
+        stop: list[str] | None = None,
+        **kwargs: Any,
+    ) -> dict:
+        payload = {**self._default_params, **kwargs}
+
+        if self._use_responses_api(payload):
+            return super()._get_request_payload(input_, stop=stop, **kwargs)
+
+        messages = self._convert_input(input_).to_messages()
+        if stop is not None:
+            kwargs["stop"] = stop
+
+        payload_messages = []
+        last_human_index = -1
+        if self.reasoning_keep_policy == "current":
+            last_human_index = _get_last_human_message_index(messages)
+
+        for index, m in enumerate(messages):
+            if isinstance(m, AIMessage):
+                msg_dict = _convert_message_to_dict(
+                    _convert_from_v1_to_chat_completions(m)
+                )
+                if self.reasoning_keep_policy == "all" and m.additional_kwargs.get(
+                    "reasoning_content"
+                ):
+                    msg_dict["reasoning_content"] = m.additional_kwargs.get(
+                        "reasoning_content"
+                    )
+                elif (
+                    self.reasoning_keep_policy == "current"
+                    and index > last_human_index
+                    and m.additional_kwargs.get("reasoning_content")
+                ):
+                    msg_dict["reasoning_content"] = m.additional_kwargs.get(
+                        "reasoning_content"
+                    )
+                payload_messages.append(msg_dict)
+            else:
+                payload_messages.append(_convert_message_to_dict(m))
+
+        payload["messages"] = payload_messages
+        return payload
+
+    @model_validator(mode="after")
+    def validate_environment(self) -> Self:
+        if not (self.api_key and self.api_key.get_secret_value()):
+            msg = f"{self._provider.upper()}_API_KEY must be set."
+            raise ValueError(msg)
+        client_params: dict = {
+            k: v
+            for k, v in {
+                "api_key": self.api_key.get_secret_value() if self.api_key else None,
+                "base_url": self.api_base,
+                "timeout": self.request_timeout,
+                "max_retries": self.max_retries,
+                "default_headers": self.default_headers,
+                "default_query": self.default_query,
+            }.items()
+            if v is not None
+        }
+
+        if not (self.client or None):
+            sync_specific: dict = {"http_client": self.http_client}
+            self.root_client = openai.OpenAI(**client_params, **sync_specific)
+            self.client = self.root_client.chat.completions
+        if not (self.async_client or None):
+            async_specific: dict = {"http_client": self.http_async_client}
+            self.root_async_client = openai.AsyncOpenAI(
+                **client_params,
+                **async_specific,
+            )
+            self.async_client = self.root_async_client.chat.completions
+        return self
+
+    @model_validator(mode="after")
+    def _set_model_profile(self) -> Self:
+        """Set model profile if not overridden."""
+        if self.profile is None:
+            self.profile = {}
+        return self
+
+    def _create_chat_result(
+        self,
+        response: Union[dict, openai.BaseModel],
+        generation_info: Optional[dict] = None,
+    ) -> ChatResult:
+        """Convert API response to LangChain ChatResult with enhanced content processing.
+
+        Extends base implementation to capture and preserve reasoning content from
+        model responses, supporting advanced models that provide reasoning chains
+        or thought processes alongside regular responses.
+
+        Handles multiple response formats:
+        - Standard OpenAI response objects with `reasoning_content` attribute
+        - Responses with `model_extra` containing reasoning data
+        - Dictionary responses (pass-through to base implementation)
+
+        Args:
+            response: Raw API response (OpenAI object or dict)
+            generation_info: Additional generation metadata
+
+        Returns:
+            ChatResult with enhanced message containing reasoning content when available
+        """
+        rtn = super()._create_chat_result(response, generation_info)
+
+        if not isinstance(response, openai.BaseModel):
+            return rtn
+
+        for generation in rtn.generations:
+            if generation.message.response_metadata is None:
+                generation.message.response_metadata = {}
+            generation.message.response_metadata["model_provider"] = "openai-compatible"
+
+        choices = getattr(response, "choices", None)
+        if choices and hasattr(choices[0].message, "reasoning_content"):
+            rtn.generations[0].message.additional_kwargs["reasoning_content"] = choices[
+                0
+            ].message.reasoning_content
+        elif choices and hasattr(choices[0].message, "model_extra"):
+            model_extra = choices[0].message.model_extra
+            if isinstance(model_extra, dict) and (
+                reasoning := model_extra.get("reasoning")
+            ):
+                rtn.generations[0].message.additional_kwargs["reasoning_content"] = (
+                    reasoning
+                )
+
+        return rtn
+
+    def _convert_chunk_to_generation_chunk(
+        self,
+        chunk: dict,
+        default_chunk_class: type,
+        base_generation_info: Optional[dict],
+    ) -> Optional[ChatGenerationChunk]:
+        """Convert streaming chunk to generation chunk with reasoning content support.
+
+        Processes streaming response chunks to extract reasoning content alongside
+        regular message content, enabling real-time streaming of both response
+        text and reasoning chains from compatible models.
+
+        Args:
+            chunk: Raw streaming chunk from API
+            default_chunk_class: Expected chunk type for validation
+            base_generation_info: Base metadata for the generation
+
+        Returns:
+            ChatGenerationChunk with reasoning content when present in chunk data
+        """
+        generation_chunk = super()._convert_chunk_to_generation_chunk(
+            chunk,
+            default_chunk_class,
+            base_generation_info,
+        )
+        if (choices := chunk.get("choices")) and generation_chunk:
+            top = choices[0]
+            if isinstance(generation_chunk.message, AIMessageChunk):
+                generation_chunk.message.response_metadata = {
+                    **generation_chunk.message.response_metadata,
+                    "model_provider": "openai-compatible",
+                }
+                if (
+                    reasoning_content := top.get("delta", {}).get("reasoning_content")
+                ) is not None:
+                    generation_chunk.message.additional_kwargs["reasoning_content"] = (
+                        reasoning_content
+                    )
+                elif (reasoning := top.get("delta", {}).get("reasoning")) is not None:
+                    generation_chunk.message.additional_kwargs["reasoning_content"] = (
+                        reasoning
+                    )
+
+        return generation_chunk
+
+    def _stream(
+        self,
+        messages: List[BaseMessage],
+        stop: Optional[List[str]] = None,
+        run_manager: Optional[CallbackManagerForLLMRun] = None,
+        **kwargs: Any,
+    ) -> Iterator[ChatGenerationChunk]:
+        if self._use_responses_api({**kwargs, **self.model_kwargs}):
+            for chunk in super()._stream_responses(
+                messages, stop=stop, run_manager=run_manager, **kwargs
+            ):
+                yield chunk
+        else:
+            if self.include_usage:
+                kwargs["stream_options"] = {"include_usage": True}
+            try:
+                for chunk in super()._stream(
+                    messages, stop=stop, run_manager=run_manager, **kwargs
+                ):
+                    yield chunk
+            except JSONDecodeError as e:
+                raise JSONDecodeError(
+                    f"{self._provider.title()} API returned an invalid response. "
+                    "Please check the API status and try again.",
+                    e.doc,
+                    e.pos,
+                ) from e
+
+    async def _astream(
+        self,
+        messages: List[BaseMessage],
+        stop: Optional[List[str]] = None,
+        run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
+        **kwargs: Any,
+    ) -> AsyncIterator[ChatGenerationChunk]:
+        if self._use_responses_api({**kwargs, **self.model_kwargs}):
+            async for chunk in super()._astream_responses(
+                messages, stop=stop, run_manager=run_manager, **kwargs
+            ):
+                yield chunk
+        else:
+            if self.include_usage:
+                kwargs["stream_options"] = {"include_usage": True}
+            try:
+                async for chunk in super()._astream(
+                    messages, stop=stop, run_manager=run_manager, **kwargs
+                ):
+                    yield chunk
+            except JSONDecodeError as e:
+                raise JSONDecodeError(
+                    f"{self._provider.title()} API returned an invalid response. "
+                    "Please check the API status and try again.",
+                    e.doc,
+                    e.pos,
+                ) from e
+
+    def _generate(
+        self,
+        messages: List[BaseMessage],
+        stop: Optional[List[str]] = None,
+        run_manager: Optional[CallbackManagerForLLMRun] = None,
+        **kwargs: Any,
+    ) -> ChatResult:
+        try:
+            return super()._generate(
+                messages, stop=stop, run_manager=run_manager, **kwargs
+            )
+        except JSONDecodeError as e:
+            raise JSONDecodeError(
+                f"{self._provider.title()} API returned an invalid response. "
+                "Please check the API status and try again.",
+                e.doc,
+                e.pos,
+            ) from e
+
+    async def _agenerate(
+        self,
+        messages: List[BaseMessage],
+        stop: Optional[List[str]] = None,
+        run_manager: Optional[AsyncCallbackManagerForLLMRun] = None,
+        **kwargs: Any,
+    ) -> ChatResult:
+        try:
+            return await super()._agenerate(
+                messages, stop=stop, run_manager=run_manager, **kwargs
+            )
+        except JSONDecodeError as e:
+            raise JSONDecodeError(
+                f"{self._provider.title()} API returned an invalid response. "
+                "Please check the API status and try again.",
+                e.doc,
+                e.pos,
+            ) from e
+
+    def bind_tools(
+        self,
+        tools: Sequence[dict[str, Any] | type | Callable | BaseTool],
+        *,
+        tool_choice: dict | str | bool | None = None,
+        strict: bool | None = None,
+        parallel_tool_calls: bool | None = None,
+        **kwargs: Any,
+    ) -> Runnable[LanguageModelInput, AIMessage]:
+        if parallel_tool_calls is not None:
+            kwargs["parallel_tool_calls"] = parallel_tool_calls
+        formatted_tools = [
+            convert_to_openai_tool(tool, strict=strict) for tool in tools
+        ]
+
+        tool_names = []
+        for tool in formatted_tools:
+            if "function" in tool:
+                tool_names.append(tool["function"]["name"])
+            elif "name" in tool:
+                tool_names.append(tool["name"])
+            else:
+                pass
+
+        support_tool_choice = False
+        if tool_choice is not None:
+            if isinstance(tool_choice, bool):
+                tool_choice = "required"
+            if isinstance(tool_choice, str):
+                if (
+                    tool_choice in ["auto", "none", "required"]
+                    and tool_choice in self.supported_tool_choice
+                ):
+                    support_tool_choice = True
+
+                elif "specific" in self.supported_tool_choice:
+                    if tool_choice in tool_names:
+                        support_tool_choice = True
+                        tool_choice = {
+                            "type": "function",
+                            "function": {"name": tool_choice},
+                        }
+            tool_choice = tool_choice if support_tool_choice else None
+        if tool_choice:
+            kwargs["tool_choice"] = tool_choice
+        return super().bind(tools=formatted_tools, **kwargs)
+
+    def with_structured_output(
+        self,
+        schema: Optional[_DictOrPydanticClass] = None,
+        *,
+        method: Literal[
+            "auto",
+            "function_calling",
+            "json_mode",
+            "json_schema",
+        ] = "auto",
+        include_raw: bool = False,
+        strict: Optional[bool] = None,
+        **kwargs: Any,
+    ) -> Runnable[LanguageModelInput, _DictOrPydantic]:
+        """Configure structured output extraction with provider compatibility handling.
+
+        Enables parsing of model outputs into structured formats (Pydantic models
+        or dictionaries) while handling provider-specific method compatibility.
+        Falls back from json_schema to function_calling for providers that don't
+        support the json_schema method.
+
+        Args:
+            schema: Output schema (Pydantic model class or dictionary definition)
+            method: Extraction method - defaults to auto,it will choice best method based on provider supported response format
+            include_raw: Whether to include raw model response alongside parsed output
+            strict: Schema enforcement strictness (provider-dependent)
+            **kwargs: Additional structured output parameters
+
+        Returns:
+            Runnable configured for structured output extraction
+        """
+        if method not in ["auto", "function_calling", "json_mode", "json_schema"]:
+            raise ValueError(
+                f"Unsupported method: {method}. Please choose from 'auto', 'function_calling', 'json_mode', 'json_schema'."
+            )
+        if method == "auto":
+            if "json_schema" in self.supported_response_format:
+                method = "json_schema"
+            else:
+                method = "function_calling"
+        elif (
+            method == "json_schema"
+            and "json_schema" not in self.supported_response_format
+        ):
+            method = "function_calling"
+        elif (
+            method == "json_mode" and "json_mode" not in self.supported_response_format
+        ):
+            method = "function_calling"
+
+        return super().with_structured_output(
+            schema,
+            method=method,
+            include_raw=include_raw,
+            strict=strict,
+            **kwargs,
+        )
+
+
+def _create_openai_compatible_model(
+    provider: str,
+    base_url: str,
+    compatibility_options: Optional[CompatibilityOptions] = None,
+) -> Type[_BaseChatOpenAICompatible]:
+    """Factory function for creating provider-specific OpenAI-compatible model classes.
+
+    Dynamically generates model classes for different OpenAI-compatible providers,
+    configuring environment variable mappings and default base URLs specific to each provider.
+
+    Args:
+        provider: Provider identifier (e.g., `vllm`,`openrouter`)
+        base_url: Default API base URL for the provider
+        compatibility_options: Optional configuration for the provider
+
+    Returns:
+        Configured model class ready for instantiation with provider-specific settings
+    """
+    chat_model_cls_name = f"Chat{provider.title()}"
+    if compatibility_options is None:
+        compatibility_options = {}
+
+    return create_model(
+        chat_model_cls_name,
+        __base__=_BaseChatOpenAICompatible,
+        api_base=(
+            str,
+            Field(
+                default_factory=from_env(
+                    f"{provider.upper()}_API_BASE", default=base_url
+                ),
+            ),
+        ),
+        api_key=(
+            str,
+            Field(
+                default_factory=secret_from_env(
+                    f"{provider.upper()}_API_KEY", default=None
+                ),
+            ),
+        ),
+        _provider=(
+            str,
+            PrivateAttr(default=provider),
+        ),
+        supported_tool_choice=(
+            ToolChoiceType,
+            Field(default=compatibility_options.get("supported_tool_choice", ["auto"])),
+        ),
+        reasoning_keep_policy=(
+            ReasoningKeepPolicy,
+            Field(default=compatibility_options.get("reasoning_keep_policy", "never")),
+        ),
+        supported_response_format=(
+            ResponseFormatType,
+            Field(default=compatibility_options.get("supported_response_format", [])),
+        ),
+        include_usage=(
+            bool,
+            Field(default=compatibility_options.get("include_usage", True)),
+        ),
+    )

{langchain_dev_utils-1.2.9.dist-info → langchain_dev_utils-1.2.10.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: langchain-dev-utils
-Version: 1.2.9
+Version: 1.2.10
 Summary: A practical utility library for LangChain and LangGraph development
 Project-URL: Source Code, https://github.com/TBice123123/langchain-dev-utils
 Project-URL: repository, https://github.com/TBice123123/langchain-dev-utils
@@ -208,16 +208,20 @@ import datetime
 from langchain_dev_utils.agents import create_agent, wrap_agent_as_tool
 from langchain.agents import AgentState
 
-
 @tool
 def get_current_time() -> str:
-    """Get the current time."""
+    """Get the current time"""
     return datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S")
 
+time_agent = create_agent("vllm:qwen3-4b", tools=[get_current_time], name="time-agent")
+call_time_agent_tool = wrap_agent_as_tool(time_agent)  
 
-agent = create_agent("vllm:qwen3-4b", tools=[get_current_time], name="time-agent")
-call_time_agent_tool = wrap_agent_as_tool(agent)
-response = call_time_agent_tool.invoke(
+agent = create_agent(
+    "vllm:qwen3-4b",
+    name="agent",
+    tools=[call_time_agent_tool],
+)
+response = agent.invoke(
     {"messages": [{"role": "user", "content": "What time is it now?"}]}
 )
 print(response)

{langchain_dev_utils-1.2.9.dist-info → langchain_dev_utils-1.2.10.dist-info}/RECORD

@@ -1,4 +1,4 @@
-langchain_dev_utils/__init__.py,sha256=Oh3Y6CIypkhAjW-aquBTyP3_cA-gKgKTwq9EpcWpjps,22
+langchain_dev_utils/__init__.py,sha256=GI4rgymQsPWdk2_d96NgmZBRuFM6yOZB-kysnrjBjVo,23
 langchain_dev_utils/_utils.py,sha256=MFEzR1BjXMj6HEVwt2x2omttFuDJ_rYAEbNqe99r9pM,1338
 langchain_dev_utils/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 langchain_dev_utils/agents/__init__.py,sha256=PJ-lSDZv_AXMYA3H4fx-HzJa14tPbkGmq1HX8LNfaPo,125
@@ -18,7 +18,7 @@ langchain_dev_utils/chat_models/__init__.py,sha256=YSLUyHrWEEj4y4DtGFCOnDW02VIYZ
 langchain_dev_utils/chat_models/base.py,sha256=CVMfgqMRnIKv8z4babusa2c4RKVuiWTL39mPD8cHAf4,11880
 langchain_dev_utils/chat_models/types.py,sha256=kVLbT-IbvNtWPVmyVmh58le5r8XCqrEwuFB9-TWCBJk,672
 langchain_dev_utils/chat_models/adapters/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-langchain_dev_utils/chat_models/adapters/openai_compatible.py,sha256=YgC8ups0owOVmY-Fwi7oxiktsC9LG6UZka8XNTCBW9g,21457
+langchain_dev_utils/chat_models/adapters/openai_compatible.py,sha256=v9ewQFn75Fl6urtVo4fZgg2RRrzRXGluKf-egYAI5A0,22658
 langchain_dev_utils/embeddings/__init__.py,sha256=zbEOaV86TUi9Zrg_dH9dpdgacWg31HMJTlTQknA9EKk,244
 langchain_dev_utils/embeddings/base.py,sha256=BGoWY0L7nG9iRV3d4sSagXhECXrwvS1xA-A_OVltn3k,9406
 langchain_dev_utils/message_convert/__init__.py,sha256=ZGrHGXPKMrZ_p9MqfIVZ4jgbEyb7aC4Q7X-muuThIYU,457
@@ -31,7 +31,7 @@ langchain_dev_utils/pipeline/types.py,sha256=T3aROKKXeWvd0jcH5XkgMDQfEkLfPaiOhhV
 langchain_dev_utils/tool_calling/__init__.py,sha256=mu_WxKMcu6RoTf4vkTPbA1WSBSNc6YIqyBtOQ6iVQj4,322
 langchain_dev_utils/tool_calling/human_in_the_loop.py,sha256=7Z_QO5OZUR6K8nLoIcafc6osnvX2IYNorOJcbx6bVso,9672
 langchain_dev_utils/tool_calling/utils.py,sha256=S4-KXQ8jWmpGTXYZitovF8rxKpaSSUkFruM8LDwvcvE,2765
-langchain_dev_utils-1.2.9.dist-info/METADATA,sha256=zzBTgY8EUmuz08ofh1t1letaKywRXuqQgfPdqWBVw2Q,13279
-langchain_dev_utils-1.2.9.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
-langchain_dev_utils-1.2.9.dist-info/licenses/LICENSE,sha256=AWAOzNEcsvCEzHOF0qby5OKxviVH_eT9Yce1sgJTico,1084
-langchain_dev_utils-1.2.9.dist-info/RECORD,,
+langchain_dev_utils-1.2.10.dist-info/METADATA,sha256=2KpH7VEYB7v6uyWHJSvq0hyfNSLxlSsIZNueFbCIv44,13372
+langchain_dev_utils-1.2.10.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
+langchain_dev_utils-1.2.10.dist-info/licenses/LICENSE,sha256=AWAOzNEcsvCEzHOF0qby5OKxviVH_eT9Yce1sgJTico,1084
+langchain_dev_utils-1.2.10.dist-info/RECORD,,