langchain-google-genai 2.1.0__tar.gz → 2.1.2__tar.gz

{langchain_google_genai-2.1.0 → langchain_google_genai-2.1.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: langchain-google-genai
-Version: 2.1.0
+Version: 2.1.2
 Summary: An integration package connecting Google's genai package and LangChain
 Home-page: https://github.com/langchain-ai/langchain-google
 License: MIT
@@ -13,7 +13,7 @@ Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Requires-Dist: filetype (>=1.2.0,<2.0.0)
 Requires-Dist: google-ai-generativelanguage (>=0.6.16,<0.7.0)
-Requires-Dist: langchain-core (>=0.3.43,<0.4.0)
+Requires-Dist: langchain-core (>=0.3.49,<0.4.0)
 Requires-Dist: pydantic (>=2,<3)
 Project-URL: Repository, https://github.com/langchain-ai/langchain-google
 Project-URL: Source Code, https://github.com/langchain-ai/langchain-google/tree/main/libs/genai
@@ -77,7 +77,44 @@ The value of `image_url` can be any of the following:
 - An accessible gcs file (e.g., "gcs://path/to/file.png")
 - A base64 encoded image (e.g., `data:image/png;base64,abcd124`)
 
+#### Multimodal outputs
 
+Gemini 2.0 Flash Experimental model supports text output with inline images
+
+```
+from langchain_google_genai import ChatGoogleGenerativeAI
+
+llm = ChatGoogleGenerativeAI(model="models/gemini-2.0-flash-exp-image-generation")
+# example
+response = llm.invoke(
+    "Generate an image of a cat and say meow",
+    generation_config=dict(response_modalities=["TEXT", "IMAGE"]),
+)
+
+# Base64 encoded binary data of the image
+image_base64 = response.content[0].get("image_url").get("url").split(",")[-1]
+meow_str = response.content[1]
+```
+
+#### Multimodal Outputs in Chains
+
+```
+from langchain_core.runnables import RunnablePassthrough
+from langchain_core.prompts import ChatPromptTemplate
+
+from langchain_google_genai import ChatGoogleGenerativeAI, Modality
+
+llm = ChatGoogleGenerativeAI(
+    model="models/gemini-2.0-flash-exp-image-generation",
+    response_modalities=[Modality.TEXT, Modality.IMAGE],
+)
+
+prompt = ChatPromptTemplate(
+    [("human", "Generate an image of {animal} and tell me the sound of the animal")]
+)
+chain = {"animal": RunnablePassthrough()} | prompt | llm
+res = chain.invoke("cat")
+```
 
 ## Embeddings
 

{langchain_google_genai-2.1.0 → langchain_google_genai-2.1.2}/README.md

@@ -56,7 +56,44 @@ The value of `image_url` can be any of the following:
 - An accessible gcs file (e.g., "gcs://path/to/file.png")
 - A base64 encoded image (e.g., `data:image/png;base64,abcd124`)
 
+#### Multimodal outputs
 
+Gemini 2.0 Flash Experimental model supports text output with inline images
+
+```
+from langchain_google_genai import ChatGoogleGenerativeAI
+
+llm = ChatGoogleGenerativeAI(model="models/gemini-2.0-flash-exp-image-generation")
+# example
+response = llm.invoke(
+    "Generate an image of a cat and say meow",
+    generation_config=dict(response_modalities=["TEXT", "IMAGE"]),
+)
+
+# Base64 encoded binary data of the image
+image_base64 = response.content[0].get("image_url").get("url").split(",")[-1]
+meow_str = response.content[1]
+```
+
+#### Multimodal Outputs in Chains
+
+```
+from langchain_core.runnables import RunnablePassthrough
+from langchain_core.prompts import ChatPromptTemplate
+
+from langchain_google_genai import ChatGoogleGenerativeAI, Modality
+
+llm = ChatGoogleGenerativeAI(
+    model="models/gemini-2.0-flash-exp-image-generation",
+    response_modalities=[Modality.TEXT, Modality.IMAGE],
+)
+
+prompt = ChatPromptTemplate(
+    [("human", "Generate an image of {animal} and tell me the sound of the animal")]
+)
+chain = {"animal": RunnablePassthrough()} | prompt | llm
+res = chain.invoke("cat")
+```
 
 ## Embeddings
 

{langchain_google_genai-2.1.0 → langchain_google_genai-2.1.2}/langchain_google_genai/__init__.py

@@ -55,7 +55,7 @@ embeddings.embed_query("hello, world!")
 ```
 """  # noqa: E501
 
-from langchain_google_genai._enums import HarmBlockThreshold, HarmCategory
+from langchain_google_genai._enums import HarmBlockThreshold, HarmCategory, Modality
 from langchain_google_genai.chat_models import ChatGoogleGenerativeAI
 from langchain_google_genai.embeddings import GoogleGenerativeAIEmbeddings
 from langchain_google_genai.genai_aqa import (
@@ -80,5 +80,6 @@ __all__ = [
     "GoogleVectorStore",
     "HarmBlockThreshold",
     "HarmCategory",
+    "Modality",
     "DoesNotExistsException",
 ]

{langchain_google_genai-2.1.0 → langchain_google_genai-2.1.2}/langchain_google_genai/_common.py

@@ -1,11 +1,11 @@
 from importlib import metadata
-from typing import Any, Dict, Optional, Tuple, TypedDict
+from typing import Any, Dict, List, Optional, Tuple, TypedDict
 
 from google.api_core.gapic_v1.client_info import ClientInfo
 from langchain_core.utils import secret_from_env
 from pydantic import BaseModel, Field, SecretStr
 
-from langchain_google_genai._enums import HarmBlockThreshold, HarmCategory
+from langchain_google_genai._enums import HarmBlockThreshold, HarmCategory, Modality
 
 
 class GoogleGenerativeAIError(Exception):
@@ -72,6 +72,9 @@ Supported examples:
             "A key-value dictionary representing additional headers for the model call"
         ),
     )
+    response_modalities: Optional[List[Modality]] = Field(
+        default=None, description=("A list of modalities of the response")
+    )
 
     safety_settings: Optional[Dict[HarmCategory, HarmBlockThreshold]] = None
     """The default safety settings to use for all generations. 

{langchain_google_genai-2.1.0 → langchain_google_genai-2.1.2}/langchain_google_genai/_enums.py

@@ -2,5 +2,6 @@ import google.ai.generativelanguage_v1beta as genai
 
 HarmBlockThreshold = genai.SafetySetting.HarmBlockThreshold
 HarmCategory = genai.HarmCategory
+Modality = genai.GenerationConfig.Modality
 
-__all__ = ["HarmBlockThreshold", "HarmCategory"]
+__all__ = ["HarmBlockThreshold", "HarmCategory", "Modality"]

{langchain_google_genai-2.1.0 → langchain_google_genai-2.1.2}/langchain_google_genai/_function_utils.py

@@ -61,18 +61,22 @@ _ALLOWED_SCHEMA_FIELDS_SET = set(_ALLOWED_SCHEMA_FIELDS)
 _FunctionDeclarationLike = Union[
     BaseTool, Type[BaseModel], gapic.FunctionDeclaration, Callable, Dict[str, Any]
 ]
+_GoogleSearchRetrievalLike = Union[
+    gapic.GoogleSearchRetrieval,
+    Dict[str, Any],
+]
 
 
 class _ToolDict(TypedDict):
     function_declarations: Sequence[_FunctionDeclarationLike]
+    google_search_retrieval: Optional[_GoogleSearchRetrievalLike]
 
 
 # Info: This means one tool=Sequence of FunctionDeclaration
 # The dict should be gapic.Tool like. {"function_declarations": [ { "name": ...}.
 # OpenAI like dict is not be accepted. {{'type': 'function', 'function': {'name': ...}
-_ToolsType = Union[
-    gapic.Tool, _ToolDict, _FunctionDeclarationLike, Sequence[_FunctionDeclarationLike]
-]
+_ToolType = Union[gapic.Tool, _ToolDict, _FunctionDeclarationLike]
+_ToolsType = Sequence[_ToolType]
 
 
 def _format_json_schema_to_gapic(schema: Dict[str, Any]) -> Dict[str, Any]:
@@ -122,7 +126,7 @@ def _format_dict_to_function_declaration(
 
 # Info: gapic.Tool means function_declarations and proto.Message.
 def convert_to_genai_function_declarations(
-    tools: Sequence[_ToolsType],
+    tools: _ToolsType,
 ) -> gapic.Tool:
     if not isinstance(tools, collections.abc.Sequence):
         logger.warning(
@@ -132,24 +136,54 @@ def convert_to_genai_function_declarations(
         tools = [tools]
     gapic_tool = gapic.Tool()
     for tool in tools:
-        if isinstance(tool, gapic.Tool):
-            gapic_tool.function_declarations.extend(tool.function_declarations)  # type: ignore[union-attr]
-        elif isinstance(tool, dict) and "function_declarations" not in tool:
-            fd = _format_to_gapic_function_declaration(tool)
-            gapic_tool.function_declarations.append(fd)
+        if any(f in gapic_tool for f in ["google_search_retrieval"]):
+            raise ValueError(
+                "Providing multiple google_search_retrieval"
+                " or mixing with function_declarations is not supported"
+            )
+        if isinstance(tool, (gapic.Tool)):
+            rt: gapic.Tool = (
+                tool if isinstance(tool, gapic.Tool) else tool._raw_tool  # type: ignore
+            )
+            if "google_search_retrieval" in rt:
+                gapic_tool.google_search_retrieval = rt.google_search_retrieval
+            if "function_declarations" in rt:
+                gapic_tool.function_declarations.extend(rt.function_declarations)
+            if "google_search" in rt:
+                gapic_tool.google_search = rt.google_search
         elif isinstance(tool, dict):
-            function_declarations = cast(_ToolDict, tool)["function_declarations"]
-            if not isinstance(function_declarations, collections.abc.Sequence):
-                raise ValueError(
-                    "function_declarations should be a list"
-                    f"got '{type(function_declarations)}'"
-                )
-            if function_declarations:
-                fds = [
-                    _format_to_gapic_function_declaration(fd)
-                    for fd in function_declarations
+            # not _ToolDictLike
+            if not any(
+                f in tool
+                for f in [
+                    "function_declarations",
+                    "google_search_retrieval",
                 ]
-                gapic_tool.function_declarations.extend(fds)
+            ):
+                fd = _format_to_gapic_function_declaration(tool)  # type: ignore[arg-type]
+                gapic_tool.function_declarations.append(fd)
+                continue
+            # _ToolDictLike
+            tool = cast(_ToolDict, tool)
+            if "function_declarations" in tool:
+                function_declarations = tool["function_declarations"]
+                if not isinstance(
+                    tool["function_declarations"], collections.abc.Sequence
+                ):
+                    raise ValueError(
+                        "function_declarations should be a list"
+                        f"got '{type(function_declarations)}'"
+                    )
+                if function_declarations:
+                    fds = [
+                        _format_to_gapic_function_declaration(fd)
+                        for fd in function_declarations
+                    ]
+                    gapic_tool.function_declarations.extend(fds)
+            if "google_search_retrieval" in tool:
+                gapic_tool.google_search_retrieval = gapic.GoogleSearchRetrieval(
+                    tool["google_search_retrieval"]
+                )
         else:
             fd = _format_to_gapic_function_declaration(tool)  # type: ignore[arg-type]
             gapic_tool.function_declarations.append(fd)

{langchain_google_genai-2.1.0 → langchain_google_genai-2.1.2}/langchain_google_genai/chat_models.py

@@ -32,6 +32,7 @@ from google.ai.generativelanguage_v1beta import (
 from google.ai.generativelanguage_v1beta.types import (
     Blob,
     Candidate,
+    CodeExecution,
     Content,
     FileData,
     FunctionCall,
@@ -45,9 +46,7 @@ from google.ai.generativelanguage_v1beta.types import (
     ToolConfig,
     VideoMetadata,
 )
-from google.ai.generativelanguage_v1beta.types import (
-    Tool as GoogleTool,
-)
+from google.ai.generativelanguage_v1beta.types import Tool as GoogleTool
 from langchain_core.callbacks.manager import (
     AsyncCallbackManagerForLLMRun,
     CallbackManagerForLLMRun,
@@ -72,7 +71,7 @@ from langchain_core.output_parsers.openai_tools import (
     parse_tool_calls,
 )
 from langchain_core.outputs import ChatGeneration, ChatGenerationChunk, ChatResult
-from langchain_core.runnables import Runnable, RunnablePassthrough
+from langchain_core.runnables import Runnable, RunnableConfig, RunnablePassthrough
 from langchain_core.tools import BaseTool
 from langchain_core.utils.function_calling import convert_to_openai_tool
 from pydantic import (
@@ -106,7 +105,10 @@ from langchain_google_genai._function_utils import (
     is_basemodel_subclass_safe,
     tool_to_dict,
 )
-from langchain_google_genai._image_utils import ImageBytesLoader
+from langchain_google_genai._image_utils import (
+    ImageBytesLoader,
+    image_bytes_to_b64_string,
+)
 
 from . import _genai_extension as genaix
 
@@ -430,7 +432,7 @@ def _parse_chat_history(
 def _parse_response_candidate(
     response_candidate: Candidate, streaming: bool = False
 ) -> AIMessage:
-    content: Union[None, str, List[str]] = None
+    content: Union[None, str, List[Union[str, dict]]] = None
     additional_kwargs = {}
     tool_calls = []
     invalid_tool_calls = []
@@ -455,6 +457,61 @@ def _parse_response_candidate(
             elif text:
                 raise Exception("Unexpected content type")
 
+        if hasattr(part, "executable_code") and part.executable_code is not None:
+            if part.executable_code.code and part.executable_code.language:
+                code_message = {
+                    "type": "executable_code",
+                    "executable_code": part.executable_code.code,
+                    "language": part.executable_code.language,
+                }
+                if not content:
+                    content = [code_message]
+                elif isinstance(content, str):
+                    content = [content, code_message]
+                elif isinstance(content, list):
+                    content.append(code_message)
+                else:
+                    raise Exception("Unexpected content type")
+
+        if (
+            hasattr(part, "code_execution_result")
+            and part.code_execution_result is not None
+        ):
+            if part.code_execution_result.output:
+                execution_result = {
+                    "type": "code_execution_result",
+                    "code_execution_result": part.code_execution_result.output,
+                }
+
+                if not content:
+                    content = [execution_result]
+                elif isinstance(content, str):
+                    content = [content, execution_result]
+                elif isinstance(content, list):
+                    content.append(execution_result)
+                else:
+                    raise Exception("Unexpected content type")
+
+        if part.inline_data.mime_type.startswith("image/"):
+            image_format = part.inline_data.mime_type[6:]
+            message = {
+                "type": "image_url",
+                "image_url": {
+                    "url": image_bytes_to_b64_string(
+                        part.inline_data.data, image_format=image_format
+                    )
+                },
+            }
+
+            if not content:
+                content = [message]
+            elif isinstance(content, str) and message:
+                content = [content, message]
+            elif isinstance(content, list) and message:
+                content.append(message)
+            elif message:
+                raise Exception("Unexpected content type")
+
         if part.function_call:
             function_call = {"name": part.function_call.name}
             # dump to match other function calling llm for now
@@ -498,6 +555,16 @@ def _parse_response_candidate(
                     )
     if content is None:
         content = ""
+    if any(isinstance(item, dict) and "executable_code" in item for item in content):
+        warnings.warn(
+            """
+        ⚠️ Warning: Output may vary each run.  
+        - 'executable_code': Always present.  
+        - 'execution_result' & 'image_url': May be absent for some queries.  
+
+        Validate before using in production.
+"""
+        )
 
     if streaming:
         return AIMessageChunk(
@@ -552,6 +619,8 @@ def _response_to_result(
         generation_info = {}
         if candidate.finish_reason:
             generation_info["finish_reason"] = candidate.finish_reason.name
+            # Add model_name in last chunk
+            generation_info["model_name"] = response.model_version
         generation_info["safety_ratings"] = [
             proto.Message.to_dict(safety_rating, use_integers_for_enums=False)
             for safety_rating in candidate.safety_ratings
@@ -713,6 +782,16 @@ class ChatGoogleGenerativeAI(_BaseGoogleGenerativeAI, BaseChatModel):
               'args': {'location': 'New York City, NY'},
               'id': '634582de-5186-4e4b-968b-f192f0a93678'}]
 
+    Use Search with Gemini 2:
+        .. code-block:: python
+
+            from google.ai.generativelanguage_v1beta.types import Tool as GenAITool
+            llm = ChatGoogleGenerativeAI(model="gemini-2.0-flash-exp")
+            resp = llm.invoke(
+                "When is the next total solar eclipse in US?",
+                tools=[GenAITool(google_search={})],
+            )
+
     Structured output:
         .. code-block:: python
 
@@ -824,6 +903,14 @@ class ChatGoogleGenerativeAI(_BaseGoogleGenerativeAI, BaseChatModel):
     def _llm_type(self) -> str:
         return "chat-google-generative-ai"
 
+    @property
+    def _supports_code_execution(self) -> bool:
+        return (
+            "gemini-1.5-pro" in self.model
+            or "gemini-1.5-flash" in self.model
+            or "gemini-2" in self.model
+        )
+
     @classmethod
     def is_lc_serializable(self) -> bool:
         return True
@@ -831,8 +918,8 @@ class ChatGoogleGenerativeAI(_BaseGoogleGenerativeAI, BaseChatModel):
     @model_validator(mode="after")
     def validate_environment(self) -> Self:
         """Validates params and passes them to google-generativeai package."""
-        if self.temperature is not None and not 0 <= self.temperature <= 1:
-            raise ValueError("temperature must be in the range [0.0, 1.0]")
+        if self.temperature is not None and not 0 <= self.temperature <= 2.0:
+            raise ValueError("temperature must be in the range [0.0, 2.0]")
 
         if self.top_p is not None and not 0 <= self.top_p <= 1:
             raise ValueError("top_p must be in the range [0.0, 1.0]")
@@ -895,8 +982,45 @@ class ChatGoogleGenerativeAI(_BaseGoogleGenerativeAI, BaseChatModel):
             "top_k": self.top_k,
             "n": self.n,
             "safety_settings": self.safety_settings,
+            "response_modalities": self.response_modalities,
         }
 
+    def invoke(
+        self,
+        input: LanguageModelInput,
+        config: Optional[RunnableConfig] = None,
+        *,
+        code_execution: Optional[bool] = None,
+        stop: Optional[list[str]] = None,
+        **kwargs: Any,
+    ) -> BaseMessage:
+        """
+        Enable code execution. Supported on: gemini-1.5-pro, gemini-1.5-flash,
+        gemini-2.0-flash, and gemini-2.0-pro. When enabled, the model can execute
+        code to solve problems.
+        """
+
+        """Override invoke to add code_execution parameter."""
+
+        if code_execution is not None:
+            if not self._supports_code_execution:
+                raise ValueError(
+                    f"Code execution is only supported on Gemini 1.5 Pro, \
+                    Gemini 1.5 Flash, "
+                    f"Gemini 2.0 Flash, and Gemini 2.0 Pro models. \
+                    Current model: {self.model}"
+                )
+            if "tools" not in kwargs:
+                code_execution_tool = GoogleTool(code_execution=CodeExecution())
+                kwargs["tools"] = [code_execution_tool]
+
+            else:
+                raise ValueError(
+                    "Tools are already defined." "code_execution tool can't be defined"
+                )
+
+        return super().invoke(input, config, stop=stop, **kwargs)
+
     def _get_ls_params(
         self, stop: Optional[List[str]] = None, **kwargs: Any
     ) -> LangSmithParams:
@@ -928,6 +1052,7 @@ class ChatGoogleGenerativeAI(_BaseGoogleGenerativeAI, BaseChatModel):
                 "max_output_tokens": self.max_output_tokens,
                 "top_k": self.top_k,
                 "top_p": self.top_p,
+                "response_modalities": self.response_modalities,
             }.items()
             if v is not None
         }
@@ -1174,8 +1299,12 @@ class ChatGoogleGenerativeAI(_BaseGoogleGenerativeAI, BaseChatModel):
                 "Must specify at most one of tool_choice and tool_config, received "
                 f"both:\n\n{tool_choice=}\n\n{tool_config=}"
             )
+
         formatted_tools = None
-        if tools:
+        code_execution_tool = GoogleTool(code_execution=CodeExecution())
+        if tools == [code_execution_tool]:
+            formatted_tools = tools
+        elif tools:
             formatted_tools = [convert_to_genai_function_declarations(tools)]
         elif functions:
             formatted_tools = [convert_to_genai_function_declarations(functions)]
@@ -1201,9 +1330,20 @@ class ChatGoogleGenerativeAI(_BaseGoogleGenerativeAI, BaseChatModel):
                     f"be specified if 'tools' is specified."
                 )
                 raise ValueError(msg)
-            all_names = [
-                f.name for t in formatted_tools for f in t.function_declarations
-            ]
+            all_names: List[str] = []
+            for t in formatted_tools:
+                if hasattr(t, "function_declarations"):
+                    t_with_declarations = cast(Any, t)
+                    all_names.extend(
+                        f.name for f in t_with_declarations.function_declarations
+                    )
+                elif isinstance(t, GoogleTool) and hasattr(t, "code_execution"):
+                    continue
+                else:
+                    raise TypeError(
+                        f"Tool {t} doesn't have function_declarations attribute"
+                    )
+
             tool_config = _tool_choice_to_tool_config(tool_choice, all_names)
 
         formatted_tool_config = None
@@ -1256,6 +1396,8 @@ class ChatGoogleGenerativeAI(_BaseGoogleGenerativeAI, BaseChatModel):
         include_raw: bool = False,
         **kwargs: Any,
     ) -> Runnable[LanguageModelInput, Union[Dict, BaseModel]]:
+        _ = kwargs.pop("method", None)
+        _ = kwargs.pop("strict", None)
         if kwargs:
             raise ValueError(f"Received unsupported arguments {kwargs}")
         tool_name = _get_tool_name(schema)  # type: ignore[arg-type]

{langchain_google_genai-2.1.0 → langchain_google_genai-2.1.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "langchain-google-genai"
-version = "2.1.0"
+version = "2.1.2"
 description = "An integration package connecting Google's genai package and LangChain"
 authors = []
 readme = "README.md"
@@ -12,7 +12,7 @@ license = "MIT"
 
 [tool.poetry.dependencies]
 python = ">=3.9,<4.0"
-langchain-core = "^0.3.43"
+langchain-core = "^0.3.49"
 google-ai-generativelanguage = "^0.6.16"
 pydantic = ">=2,<3"
 filetype = "^1.2.0"
@@ -28,7 +28,7 @@ syrupy = "^4.0.2"
 pytest-watcher = "^0.3.4"
 pytest-asyncio = "^0.21.1"
 numpy = "^1.26.2"
-langchain-tests = "0.3.14"
+langchain-tests = "0.3.17"
 
 [tool.codespell]
 ignore-words-list = "rouge"