instructor 1.3.2__tar.gz → 1.3.3__tar.gz

{instructor-1.3.2 → instructor-1.3.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: instructor
-Version: 1.3.2
+Version: 1.3.3
 Summary: structured outputs for llm
 Home-page: https://github.com/jxnl/instructor
 License: MIT
@@ -20,14 +20,18 @@ Provides-Extra: groq
 Provides-Extra: litellm
 Provides-Extra: mistralai
 Provides-Extra: test-docs
+Provides-Extra: vertexai
 Requires-Dist: aiohttp (>=3.9.1,<4.0.0)
-Requires-Dist: anthropic (>=0.26.0,<0.27.0) ; extra == "anthropic" or extra == "test-docs"
+Requires-Dist: anthropic (>=0.27.0,<0.28.0) ; extra == "anthropic" or extra == "test-docs"
 Requires-Dist: cohere (>=5.1.8,<6.0.0) ; extra == "cohere" or extra == "test-docs"
 Requires-Dist: diskcache (>=5.6.3,<6.0.0) ; extra == "test-docs"
 Requires-Dist: docstring-parser (>=0.16,<0.17)
 Requires-Dist: fastapi (>=0.109.2,<0.110.0) ; extra == "test-docs"
+Requires-Dist: google-cloud-aiplatform (>=1.52.0,<2.0.0) ; extra == "vertexai"
 Requires-Dist: google-generativeai (>=0.5.4,<0.6.0) ; extra == "google-generativeai"
 Requires-Dist: groq (>=0.4.2,<0.5.0) ; extra == "groq" or extra == "test-docs"
+Requires-Dist: jiter (>=0.4.1,<0.5.0)
+Requires-Dist: jsonref (>=1.1.0,<2.0.0) ; extra == "vertexai"
 Requires-Dist: litellm (>=1.35.31,<2.0.0) ; extra == "test-docs" or extra == "litellm"
 Requires-Dist: mistralai (>=0.1.8,<0.2.0) ; extra == "test-docs" or extra == "mistralai"
 Requires-Dist: openai (>=1.1.0,<2.0.0)
@@ -184,32 +188,19 @@ import instructor
 import google.generativeai as genai
 from pydantic import BaseModel
 
+
 class User(BaseModel):
     name: str
     age: int
 
+
 # genai.configure(api_key=os.environ["API_KEY"]) # alternative API key configuration
 client = instructor.from_gemini(
     client=genai.GenerativeModel(
-        model_name="models/gemini-1.5-flash-latest", # model defaults to "gemini-pro"
+        model_name="models/gemini-1.5-flash-latest",  # model defaults to "gemini-pro"
     ),
     mode=instructor.Mode.GEMINI_JSON,
 )
-
-# note that client.chat.completions.create will also work
-resp = client.chat.completions.create(
-    messages=[
-        {
-            "role": "user",
-            "content": "Extract Jason is 25 years old.",
-        }
-    ],
-    response_model=User,
-)
-
-assert isinstance(resp, User)
-assert resp.name == "Jason"
-assert resp.age == 25
 ```
 
 ### Using Litellm
@@ -369,13 +360,13 @@ for user in user_stream:
     #> name=None age=None
     #> name=None age=None
     #> name=None age=None
-    #> name=None age=25
-    #> name=None age=25
-    #> name=None age=25
-    #> name=None age=25
-    #> name=None age=25
-    #> name=None age=25
-    #> name='John Doe' age=25
+    #> name=None age=None
+    #> name=None age=None
+    #> name='John Doe' age=None
+    #> name='John Doe' age=None
+    #> name='John Doe' age=None
+    #> name='John Doe' age=30
+    #> name='John Doe' age=30
     # name=None age=None
     # name='' age=None
     # name='John' age=None
@@ -415,8 +406,8 @@ users = client.chat.completions.create_iterable(
 
 for user in users:
     print(user)
-    #> name='John' age=30
-    #> name='Jane' age=25
+    #> name='John Doe' age=30
+    #> name='Jane Doe' age=28
     # User(name='John Doe', age=30)
     # User(name='Jane Smith', age=25)
 ```

{instructor-1.3.2 → instructor-1.3.3}/README.md

@@ -138,32 +138,19 @@ import instructor
 import google.generativeai as genai
 from pydantic import BaseModel
 
+
 class User(BaseModel):
     name: str
     age: int
 
+
 # genai.configure(api_key=os.environ["API_KEY"]) # alternative API key configuration
 client = instructor.from_gemini(
     client=genai.GenerativeModel(
-        model_name="models/gemini-1.5-flash-latest", # model defaults to "gemini-pro"
+        model_name="models/gemini-1.5-flash-latest",  # model defaults to "gemini-pro"
     ),
     mode=instructor.Mode.GEMINI_JSON,
 )
-
-# note that client.chat.completions.create will also work
-resp = client.chat.completions.create(
-    messages=[
-        {
-            "role": "user",
-            "content": "Extract Jason is 25 years old.",
-        }
-    ],
-    response_model=User,
-)
-
-assert isinstance(resp, User)
-assert resp.name == "Jason"
-assert resp.age == 25
 ```
 
 ### Using Litellm
@@ -323,13 +310,13 @@ for user in user_stream:
     #> name=None age=None
     #> name=None age=None
     #> name=None age=None
-    #> name=None age=25
-    #> name=None age=25
-    #> name=None age=25
-    #> name=None age=25
-    #> name=None age=25
-    #> name=None age=25
-    #> name='John Doe' age=25
+    #> name=None age=None
+    #> name=None age=None
+    #> name='John Doe' age=None
+    #> name='John Doe' age=None
+    #> name='John Doe' age=None
+    #> name='John Doe' age=30
+    #> name='John Doe' age=30
     # name=None age=None
     # name='' age=None
     # name='John' age=None
@@ -369,8 +356,8 @@ users = client.chat.completions.create_iterable(
 
 for user in users:
     print(user)
-    #> name='John' age=30
-    #> name='Jane' age=25
+    #> name='John Doe' age=30
+    #> name='Jane Doe' age=28
     # User(name='John Doe', age=30)
     # User(name='Jane Smith', age=25)
 ```

{instructor-1.3.2 → instructor-1.3.3}/instructor/__init__.py

@@ -74,3 +74,8 @@ if importlib.util.find_spec("cohere") is not None:
     from .client_cohere import from_cohere
 
     __all__ += ["from_cohere"]
+
+if importlib.util.find_spec("vertexai") is not None:
+    from .client_vertexai import from_vertexai
+
+    __all__ += ["from_vertexai"]

{instructor-1.3.2 → instructor-1.3.3}/instructor/cli/usage.py

@@ -118,11 +118,11 @@ def calculate_cost(
 
 def group_and_sum_by_date_and_snapshot(usage_data: list[dict[str, Any]]) -> Table:
     """Group and sum the usage data by date and snapshot, including costs."""
-    summary: defaultdict[
-        str, defaultdict[str, dict[str, Union[int, float]]]
-    ] = defaultdict(
-        lambda: defaultdict(
-            lambda: {"total_requests": 0, "total_tokens": 0, "total_cost": 0.0}
+    summary: defaultdict[str, defaultdict[str, dict[str, Union[int, float]]]] = (
+        defaultdict(
+            lambda: defaultdict(
+                lambda: {"total_requests": 0, "total_tokens": 0, "total_cost": 0.0}
+            )
         )
     )
 

{instructor-1.3.2 → instructor-1.3.3}/instructor/client.py

@@ -63,8 +63,7 @@ class Instructor:
         validation_context: dict[str, Any] | None = None,
         strict: bool = True,
         **kwargs: Any,
-    ) -> Awaitable[T]:
-        ...
+    ) -> Awaitable[T]: ...
 
     @overload
     def create(
@@ -75,8 +74,7 @@ class Instructor:
         validation_context: dict[str, Any] | None = None,
         strict: bool = True,
         **kwargs: Any,
-    ) -> T:
-        ...
+    ) -> T: ...
 
     # TODO: we should overload a case where response_model is None
     def create(
@@ -108,8 +106,7 @@ class Instructor:
         validation_context: dict[str, Any] | None = None,
         strict: bool = True,
         **kwargs: Any,
-    ) -> AsyncGenerator[T, None]:
-        ...
+    ) -> AsyncGenerator[T, None]: ...
 
     @overload
     def create_partial(
@@ -120,8 +117,7 @@ class Instructor:
         validation_context: dict[str, Any] | None = None,
         strict: bool = True,
         **kwargs: Any,
-    ) -> Generator[T, None, None]:
-        ...
+    ) -> Generator[T, None, None]: ...
 
     def create_partial(
         self,
@@ -155,8 +151,7 @@ class Instructor:
         validation_context: dict[str, Any] | None = None,
         strict: bool = True,
         **kwargs: Any,
-    ) -> AsyncGenerator[T, None]:
-        ...
+    ) -> AsyncGenerator[T, None]: ...
 
     @overload
     def create_iterable(
@@ -167,8 +162,7 @@ class Instructor:
         validation_context: dict[str, Any] | None = None,
         strict: bool = True,
         **kwargs: Any,
-    ) -> Generator[T, None, None]:
-        ...
+    ) -> Generator[T, None, None]: ...
 
     def create_iterable(
         self,
@@ -203,8 +197,7 @@ class Instructor:
         validation_context: dict[str, Any] | None = None,
         strict: bool = True,
         **kwargs: Any,
-    ) -> Awaitable[tuple[T, Any]]:
-        ...
+    ) -> Awaitable[tuple[T, Any]]: ...
 
     @overload
     def create_with_completion(
@@ -215,8 +208,7 @@ class Instructor:
         validation_context: dict[str, Any] | None = None,
         strict: bool = True,
         **kwargs: Any,
-    ) -> tuple[T, Any]:
-        ...
+    ) -> tuple[T, Any]: ...
 
     def create_with_completion(
         self,
@@ -432,8 +424,7 @@ def from_litellm(
     completion: Callable[..., Any],
     mode: instructor.Mode = instructor.Mode.TOOLS,
     **kwargs: Any,
-) -> Instructor:
-    ...
+) -> Instructor: ...
 
 
 @overload

{instructor-1.3.2 → instructor-1.3.3}/instructor/client_anthropic.py

@@ -11,10 +11,9 @@ def from_anthropic(
     client: (
         anthropic.Anthropic | anthropic.AnthropicBedrock | anthropic.AnthropicVertex
     ),
-    mode: instructor.Mode = instructor.Mode.ANTHROPIC_JSON,
+    mode: instructor.Mode = instructor.Mode.ANTHROPIC_TOOLS,
     **kwargs: Any,
-) -> instructor.Instructor:
-    ...
+) -> instructor.Instructor: ...
 
 
 @overload
@@ -24,10 +23,9 @@ def from_anthropic(
         | anthropic.AsyncAnthropicBedrock
         | anthropic.AsyncAnthropicVertex
     ),
-    mode: instructor.Mode = instructor.Mode.ANTHROPIC_JSON,
+    mode: instructor.Mode = instructor.Mode.ANTHROPIC_TOOLS,
     **kwargs: Any,
-) -> instructor.AsyncInstructor:
-    ...
+) -> instructor.AsyncInstructor: ...
 
 
 def from_anthropic(
@@ -39,7 +37,7 @@ def from_anthropic(
         | anthropic.AsyncAnthropicVertex
         | anthropic.AnthropicVertex
     ),
-    mode: instructor.Mode = instructor.Mode.ANTHROPIC_JSON,
+    mode: instructor.Mode = instructor.Mode.ANTHROPIC_TOOLS,
     **kwargs: Any,
 ) -> instructor.Instructor | instructor.AsyncInstructor:
     assert (
@@ -62,10 +60,7 @@ def from_anthropic(
         ),
     ), "Client must be an instance of {anthropic.Anthropic, anthropic.AsyncAnthropic, anthropic.AnthropicBedrock, anthropic.AsyncAnthropicBedrock,  anthropic.AnthropicVertex, anthropic.AsyncAnthropicVertex}"
 
-    if mode == instructor.Mode.ANTHROPIC_TOOLS:
-        create = client.beta.tools.messages.create  # type: ignore - unknown in stubs
-    else:
-        create = client.messages.create
+    create = client.messages.create
 
     if isinstance(
         client,

{instructor-1.3.2 → instructor-1.3.3}/instructor/client_cohere.py

@@ -23,8 +23,7 @@ def from_cohere(
     client: cohere.Client,
     mode: instructor.Mode = instructor.Mode.COHERE_TOOLS,
     **kwargs: Any,
-) -> instructor.Instructor:
-    ...
+) -> instructor.Instructor: ...
 
 
 @overload
@@ -32,8 +31,7 @@ def from_cohere(
     client: cohere.AsyncClient,
     mode: instructor.Mode = instructor.Mode.COHERE_TOOLS,
     **kwargs: Any,
-) -> instructor.AsyncInstructor:
-    ...
+) -> instructor.AsyncInstructor: ...
 
 
 def from_cohere(

instructor-1.3.3/instructor/client_gemini.py

@@ -0,0 +1,61 @@
+# type: ignore
+from __future__ import annotations
+
+from typing import Any, Literal, overload
+
+import google.generativeai as genai
+
+import instructor
+
+
+@overload
+def from_gemini(
+    client: genai.GenerativeModel,
+    mode: instructor.Mode = instructor.Mode.GEMINI_JSON,
+    use_async: Literal[True] = True,
+    **kwargs: Any,
+) -> instructor.AsyncInstructor: ...
+
+
+@overload
+def from_gemini(
+    client: genai.GenerativeModel,
+    mode: instructor.Mode = instructor.Mode.GEMINI_JSON,
+    use_async: Literal[False] = False,
+    **kwargs: Any,
+) -> instructor.Instructor: ...
+
+
+def from_gemini(
+    client: genai.GenerativeModel,
+    mode: instructor.Mode = instructor.Mode.GEMINI_JSON,
+    use_async: bool = False,
+    **kwargs: Any,
+) -> instructor.Instructor | instructor.AsyncInstructor:
+    assert (
+        mode == instructor.Mode.GEMINI_JSON
+    ), "Mode must be instructor.Mode.GEMINI_JSON"
+
+    assert isinstance(
+        client,
+        (genai.GenerativeModel),
+    ), "Client must be an instance of genai.generativemodel"
+
+    if use_async:
+        create = client.generate_content_async
+        return instructor.AsyncInstructor(
+            client=client,
+            create=instructor.patch(create=create, mode=mode),
+            provider=instructor.Provider.GEMINI,
+            mode=mode,
+            **kwargs,
+        )
+
+    create = client.generate_content
+    return instructor.Instructor(
+        client=client,
+        create=instructor.patch(create=create, mode=mode),
+        provider=instructor.Provider.GEMINI,
+        mode=mode,
+        **kwargs,
+    )

{instructor-1.3.2 → instructor-1.3.3}/instructor/client_groq.py

@@ -11,8 +11,7 @@ def from_groq(
     client: groq.Groq,
     mode: instructor.Mode = instructor.Mode.TOOLS,
     **kwargs: Any,
-) -> instructor.Instructor:
-    ...
+) -> instructor.Instructor: ...
 
 
 @overload
@@ -20,8 +19,7 @@ def from_groq(
     client: groq.AsyncGroq,
     mode: instructor.Mode = instructor.Mode.TOOLS,
     **kwargs: Any,
-) -> instructor.AsyncInstructor:
-    ...
+) -> instructor.AsyncInstructor: ...
 
 
 def from_groq(

{instructor-1.3.2 → instructor-1.3.3}/instructor/client_mistral.py

@@ -12,8 +12,7 @@ def from_mistral(
     client: mistralai.client.MistralClient,
     mode: instructor.Mode = instructor.Mode.MISTRAL_TOOLS,
     **kwargs: Any,
-) -> instructor.Instructor:
-    ...
+) -> instructor.Instructor: ...
 
 
 @overload
@@ -21,8 +20,7 @@ def from_mistral(
     client: mistralaiasynccli.MistralAsyncClient,
     mode: instructor.Mode = instructor.Mode.MISTRAL_TOOLS,
     **kwargs: Any,
-) -> instructor.AsyncInstructor:
-    ...
+) -> instructor.AsyncInstructor: ...
 
 
 def from_mistral(

instructor-1.3.3/instructor/client_vertexai.py

@@ -0,0 +1,90 @@
+from __future__ import annotations
+
+from typing import Any
+
+from vertexai.preview.generative_models import ToolConfig  # type: ignore[reportMissingTypeStubs]
+import vertexai.generative_models as gm  # type: ignore[reportMissingTypeStubs]
+from pydantic import BaseModel
+import instructor
+import jsonref  # type: ignore[reportMissingTypeStubs]
+
+
+def _create_vertexai_tool(model: BaseModel) -> gm.Tool:
+    schema: dict[Any, Any] = jsonref.replace_refs(model.model_json_schema())  # type: ignore[reportMissingTypeStubs]
+
+    parameters: dict[Any, Any] = {
+        "type": schema["type"],
+        "properties": schema["properties"],
+        "required": schema["required"],
+    }
+
+    declaration = gm.FunctionDeclaration(
+        name=model.__name__, description=model.__doc__, parameters=parameters
+    )
+
+    tool = gm.Tool(function_declarations=[declaration])
+
+    return tool
+
+
+def _vertexai_message_parser(message: dict[str, str]) -> gm.Content:
+    return gm.Content(
+        role=message["role"], parts=[gm.Part.from_text(message["content"])]
+    )
+
+
+def vertexai_function_response_parser(
+    response: gm.GenerationResponse, exception: Exception
+) -> gm.Content:
+    return gm.Content(
+        parts=[
+            gm.Part.from_function_response(
+                name=response.candidates[0].content.parts[0].function_call.name,
+                response={
+                    "content": f"Validation Error found:\n{exception}\nRecall the function correctly, fix the errors"
+                },
+            )
+        ]
+    )
+
+
+def vertexai_process_response(_kwargs: dict[str, Any], model: BaseModel):
+    messages = _kwargs.pop("messages")
+    contents = [
+        _vertexai_message_parser(message)  # type: ignore[reportUnkownArgumentType]
+        if isinstance(message, dict)
+        else message
+        for message in messages
+    ]
+    tool = _create_vertexai_tool(model=model)
+    tool_config = ToolConfig(
+        function_calling_config=ToolConfig.FunctionCallingConfig(
+            mode=ToolConfig.FunctionCallingConfig.Mode.ANY,
+        )
+    )
+    return contents, [tool], tool_config
+
+
+def from_vertexai(
+    client: gm.GenerativeModel,
+    mode: instructor.Mode = instructor.Mode.VERTEXAI_TOOLS,
+    _async: bool = False,
+    **kwargs: Any,
+) -> instructor.Instructor:
+    assert (
+        mode == instructor.Mode.VERTEXAI_TOOLS
+    ), "Mode must be instructor.Mode.VERTEXAI_TOOLS"
+
+    assert isinstance(
+        client, gm.GenerativeModel
+    ), "Client must be an instance of vertexai.generative_models.GenerativeModel"
+
+    create = client.generate_content_async if _async else client.generate_content
+
+    return instructor.Instructor(
+        client=client,
+        create=instructor.patch(create=create, mode=mode),
+        provider=instructor.Provider.VERTEXAI,
+        mode=mode,
+        **kwargs,
+    )

{instructor-1.3.2 → instructor-1.3.3}/instructor/dsl/parallel.py

@@ -1,3 +1,4 @@
+import sys
 from typing import (
     Any,
     Optional,
@@ -44,18 +45,24 @@ class ParallelBase:
             )
 
 
+if sys.version_info >= (3, 10):
+    from types import UnionType
+
+    def is_union_type(typehint: type[Iterable[T]]) -> bool:
+        return get_origin(get_args(typehint)[0]) in (Union, UnionType)
+else:
+
+    def is_union_type(typehint: type[Iterable[T]]) -> bool:
+        return get_origin(get_args(typehint)[0]) is Union
+
+
 def get_types_array(typehint: type[Iterable[T]]) -> tuple[type[T], ...]:
     should_be_iterable = get_origin(typehint)
     if should_be_iterable is not Iterable:
         raise TypeError(f"Model should be with Iterable instead if {typehint}")
 
-    if get_origin(get_args(typehint)[0]) is Union:
-        # works for Iterable[Union[int, str]]
-        the_types = get_args(get_args(typehint)[0])
-        return the_types
-
-    if get_origin(get_args(typehint)[0]) is Union:
-        # works for Iterable[Union[int, str]]
+    if is_union_type(typehint):
+        # works for Iterable[Union[int, str]], Iterable[int | str]
         the_types = get_args(get_args(typehint)[0])
         return the_types
 

{instructor-1.3.2 → instructor-1.3.3}/instructor/dsl/partial.py

@@ -8,7 +8,7 @@
 
 from __future__ import annotations
 
-import pydantic_core
+from jiter import from_json
 from pydantic import BaseModel, create_model  # type: ignore - remove once Pydantic is updated
 from pydantic.fields import FieldInfo
 from typing import (
@@ -127,8 +127,9 @@ class PartialBase(Generic[T_Model]):
         partial_model = cls.get_partial_model()
         for chunk in json_chunks:
             potential_object += chunk
-
-            obj = pydantic_core.from_json(potential_object or "{}", allow_partial=True)
+            obj = from_json(
+                (potential_object or "{}").encode(), partial_mode="trailing-strings"
+            )
             obj = partial_model.model_validate(obj, strict=None, **kwargs)
             yield obj
 
@@ -140,7 +141,9 @@ class PartialBase(Generic[T_Model]):
         partial_model = cls.get_partial_model()
         async for chunk in json_chunks:
             potential_object += chunk
-            obj = pydantic_core.from_json(potential_object or "{}", allow_partial=True)
+            obj = from_json(
+                (potential_object or "{}").encode(), partial_mode="trailing-strings"
+            )
             obj = partial_model.model_validate(obj, strict=None, **kwargs)
             yield obj
 

{instructor-1.3.2 → instructor-1.3.3}/instructor/dsl/simple_type.py

@@ -23,21 +23,41 @@ class ModelAdapter(typing.Generic[T]):
 
     def __class_getitem__(cls, response_model: type[BaseModel]) -> type[BaseModel]:
         assert is_simple_type(response_model), "Only simple types are supported"
-        tmp = create_model(
+        return create_model(
             "Response",
             content=(response_model, ...),
             __doc__="Correctly Formated and Extracted Response.",
             __base__=(AdapterBase, OpenAISchema),
         )
-        return tmp
+
+
+def validateIsSubClass(response_model: type):
+    """
+    Temporary guard against issues with generics in Python 3.9
+    """
+    import sys
+
+    if sys.version_info < (3, 10):
+        if len(typing.get_args(response_model)) == 0:
+            return False
+        return issubclass(typing.get_args(response_model)[0], BaseModel)
+    return issubclass(response_model, BaseModel)
 
 
 def is_simple_type(
-    response_model: type[BaseModel] | str | int | float | bool,
+    response_model: type[BaseModel] | str | int | float | bool | typing.Any,
 ) -> bool:
     # ! we're getting mixes between classes and instances due to how we handle some
     # ! response model types, we should fix this in later PRs
-    if isclass(response_model) and issubclass(response_model, BaseModel):
+
+    try:
+        if isclass(response_model) and validateIsSubClass(response_model):
+            return False
+    except TypeError:
+        # ! In versions < 3.11, typing.Iterable is not a class, so we can't use isclass
+        # ! for now if `response_model` is an Iterable isclass and issubclass will raise
+        # ! TypeError, so we need to check if `response_model` is an Iterable
+        # ! This is a workaround for now, we should fix this in later PRs
         return False
 
     if typing.get_origin(response_model) in {typing.Iterable, Partial}:

instructor-1.3.3/instructor/exceptions.py

@@ -0,0 +1,34 @@
+from __future__ import annotations
+
+from typing import Any
+
+
+class IncompleteOutputException(Exception):
+    """Exception raised when the output from LLM is incomplete due to max tokens limit reached."""
+
+    def __init__(
+        self,
+        *args: list[Any],
+        last_completion: Any | None = None,
+        message: str = "The output is incomplete due to a max_tokens length limit.",
+        **kwargs: dict[str, Any],
+    ):
+        self.last_completion = last_completion
+        super().__init__(message, *args, **kwargs)
+
+
+class InstructorRetryException(Exception):
+    def __init__(
+        self,
+        *args: list[Any],
+        last_completion: Any | None = None,
+        messages: list[Any] | None = None,
+        n_attempts: int,
+        total_usage: int,
+        **kwargs: dict[str, Any],
+    ):
+        self.last_completion = last_completion
+        self.messages = messages
+        self.n_attempts = n_attempts
+        self.total_usage = total_usage
+        super().__init__(*args, **kwargs)

{instructor-1.3.2 → instructor-1.3.3}/instructor/function_calls.py

@@ -1,3 +1,4 @@
+# type: ignore
 import json
 import logging
 from functools import wraps
@@ -5,7 +6,7 @@ from typing import Annotated, Any, Optional, TypeVar, cast
 
 from docstring_parser import parse
 from openai.types.chat import ChatCompletion
-from pydantic import (  # type: ignore - remove once Pydantic is updated
+from pydantic import (
     BaseModel,
     ConfigDict,
     Field,
@@ -102,6 +103,9 @@ class OpenAISchema(BaseModel):
         if mode == Mode.ANTHROPIC_JSON:
             return cls.parse_anthropic_json(completion, validation_context, strict)
 
+        if mode == Mode.VERTEXAI_TOOLS:
+            return cls.parse_vertexai_tools(completion, validation_context, strict)
+
         if mode == Mode.COHERE_TOOLS:
             return cls.parse_cohere_tools(completion, validation_context, strict)
 
@@ -165,7 +169,7 @@ class OpenAISchema(BaseModel):
     @classmethod
     def parse_gemini_json(
         cls: type[BaseModel],
-        completion: ChatCompletion,
+        completion: Any,
         validation_context: Optional[dict[str, Any]] = None,
         strict: Optional[bool] = None,
     ) -> BaseModel:
@@ -173,10 +177,13 @@ class OpenAISchema(BaseModel):
             text = completion.text
         except ValueError:
             logger.debug(
-                f"Error response: {completion._result.candidates[0].finish_reason}\n\n{completion_result.candidates[0].safety_ratings}"
+                f"Error response: {completion.result.candidates[0].finish_reason}\n\n{completion.result.candidates[0].safety_ratings}"
             )
 
-        extra_text = extract_json_from_codeblock(text)
+        try:
+            extra_text = extract_json_from_codeblock(text)  # type: ignore
+        except UnboundLocalError:
+            raise ValueError("Unable to extract JSON from completion text") from None
 
         if strict:
             return cls.model_validate_json(
@@ -188,6 +195,20 @@ class OpenAISchema(BaseModel):
             # Pydantic non-strict: https://docs.pydantic.dev/latest/concepts/strict_mode/
             return cls.model_validate(parsed, context=validation_context, strict=False)
 
+    @classmethod
+    def parse_vertexai_tools(
+        cls: type[BaseModel],
+        completion: ChatCompletion,
+        validation_context: Optional[dict[str, Any]] = None,
+        strict: Optional[bool] = None,
+    ) -> BaseModel:
+        strict = False
+        tool_call = completion.candidates[0].content.parts[0].function_call.args  # type: ignore
+        model = {}
+        for field in tool_call:  # type: ignore
+            model[field] = tool_call[field]
+        return cls.model_validate(model, context=validation_context, strict=strict)
+
     @classmethod
     def parse_cohere_tools(
         cls: type[BaseModel],

{instructor-1.3.2 → instructor-1.3.3}/instructor/mode.py

@@ -3,7 +3,7 @@ import warnings
 
 
 class _WarnOnFunctionsAccessEnumMeta(enum.EnumMeta):
-    def __getattribute__(cls, name):
+    def __getattribute__(cls, name: str):
         if name == "FUNCTIONS":
             warnings.warn(
                 "FUNCTIONS is deprecated and will be removed in future versions",
@@ -26,4 +26,5 @@ class Mode(enum.Enum, metaclass=_WarnOnFunctionsAccessEnumMeta):
     ANTHROPIC_TOOLS = "anthropic_tools"
     ANTHROPIC_JSON = "anthropic_json"
     COHERE_TOOLS = "cohere_tools"
+    VERTEXAI_TOOLS = "vertexai_tools"
     GEMINI_JSON = "gemini_json"

{instructor-1.3.2 → instructor-1.3.3}/instructor/patch.py

@@ -35,8 +35,7 @@ class InstructorChatCompletionCreate(Protocol):
         max_retries: int = 1,
         *args: T_ParamSpec.args,
         **kwargs: T_ParamSpec.kwargs,
-    ) -> T_Model:
-        ...
+    ) -> T_Model: ...
 
 
 class AsyncInstructorChatCompletionCreate(Protocol):
@@ -47,40 +46,35 @@ class AsyncInstructorChatCompletionCreate(Protocol):
         max_retries: int = 1,
         *args: T_ParamSpec.args,
         **kwargs: T_ParamSpec.kwargs,
-    ) -> T_Model:
-        ...
+    ) -> T_Model: ...
 
 
 @overload
 def patch(
     client: OpenAI,
     mode: Mode = Mode.TOOLS,
-) -> OpenAI:
-    ...
+) -> OpenAI: ...
 
 
 @overload
 def patch(
     client: AsyncOpenAI,
     mode: Mode = Mode.TOOLS,
-) -> AsyncOpenAI:
-    ...
+) -> AsyncOpenAI: ...
 
 
 @overload
 def patch(
     create: Callable[T_ParamSpec, T_Retval],
     mode: Mode = Mode.TOOLS,
-) -> InstructorChatCompletionCreate:
-    ...
+) -> InstructorChatCompletionCreate: ...
 
 
 @overload
 def patch(
     create: Awaitable[T_Retval],
     mode: Mode = Mode.TOOLS,
-) -> InstructorChatCompletionCreate:
-    ...
+) -> InstructorChatCompletionCreate: ...
 
 
 def patch(

{instructor-1.3.2 → instructor-1.3.3}/instructor/process_response.py

@@ -417,6 +417,15 @@ The output must be a valid JSON object that `{response_model.__name__}.model_val
                 HarmCategory.HARM_CATEGORY_HARASSMENT: HarmBlockThreshold.BLOCK_ONLY_HIGH,
                 HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT: HarmBlockThreshold.BLOCK_ONLY_HIGH,
             }
+        elif mode == Mode.VERTEXAI_TOOLS:
+            from instructor.client_vertexai import vertexai_process_response
+
+            contents, tools, tool_config = vertexai_process_response(
+                new_kwargs, response_model
+            )
+            new_kwargs["contents"] = contents
+            new_kwargs["tools"] = tools
+            new_kwargs["tool_config"] = tool_config
         else:
             raise ValueError(f"Invalid patch mode: {mode}")
 

{instructor-1.3.2 → instructor-1.3.3}/instructor/retry.py

@@ -11,6 +11,7 @@ from instructor.utils import (
     update_total_usage,
     merge_consecutive_messages,
 )
+from instructor.exceptions import InstructorRetryException
 
 from openai.types.completion_usage import CompletionUsage
 from pydantic import ValidationError
@@ -30,23 +31,6 @@ T_ParamSpec = ParamSpec("T_ParamSpec")
 T = TypeVar("T")
 
 
-class InstructorRetryException(Exception):
-    def __init__(
-        self,
-        *args,
-        last_completion,
-        messages: list,
-        n_attempts: int,
-        total_usage,
-        **kwargs,
-    ):
-        self.last_completion = last_completion
-        self.messages = messages
-        self.n_attempts = n_attempts
-        self.total_usage = total_usage
-        super().__init__(*args, **kwargs)
-
-
 def reask_messages(response: ChatCompletion, mode: Mode, exception: Exception):
     if mode == Mode.ANTHROPIC_TOOLS:
         # The original response
@@ -107,6 +91,12 @@ def reask_messages(response: ChatCompletion, mode: Mode, exception: Exception):
             ],
         }
         return
+    if mode == Mode.VERTEXAI_TOOLS:
+        from .client_vertexai import vertexai_function_response_parser
+
+        yield response.candidates[0].content
+        yield vertexai_function_response_parser(response, exception)
+        return
 
     yield dump_message(response.choices[0].message)
     # TODO: Give users more control on configuration
@@ -157,6 +147,7 @@ def retry_sync(
         raise ValueError("max_retries must be an int or a `tenacity.Retrying` object")
 
     try:
+        response = None
         for attempt in max_retries:
             with attempt:
                 try:
@@ -173,7 +164,7 @@ def retry_sync(
                     )
                 except (ValidationError, JSONDecodeError) as e:
                     logger.debug(f"Error response: {response}")
-                    if mode in {Mode.GEMINI_JSON}:
+                    if mode in {Mode.GEMINI_JSON, Mode.VERTEXAI_TOOLS}:
                         kwargs["contents"].extend(reask_messages(response, mode, e))
                     else:
                         kwargs["messages"].extend(reask_messages(response, mode, e))
@@ -181,13 +172,7 @@ def retry_sync(
                         kwargs["messages"] = merge_consecutive_messages(
                             kwargs["messages"]
                         )
-                    raise InstructorRetryException(
-                        e,
-                        last_completion=response,
-                        n_attempts=attempt.retry_state.attempt_number,
-                        messages=kwargs.get("messages", kwargs.get("contents")),
-                        total_usage=total_usage,
-                    ) from e
+                    raise e
     except RetryError as e:
         raise InstructorRetryException(
             e,
@@ -227,6 +212,7 @@ async def retry_async(
         )
 
     try:
+        response = None
         async for attempt in max_retries:
             logger.debug(f"Retrying, attempt: {attempt}")
             with attempt:
@@ -249,13 +235,7 @@ async def retry_async(
                         kwargs["messages"] = merge_consecutive_messages(
                             kwargs["messages"]
                         )
-                    raise InstructorRetryException(
-                        e,
-                        last_completion=response,
-                        n_attempts=attempt.retry_state.attempt_number,
-                        messages=kwargs["messages"],
-                        total_usage=total_usage,
-                    ) from e
+                    raise e
     except RetryError as e:
         logger.exception(f"Failed after retries: {e.last_attempt.exception}")
         raise InstructorRetryException(

{instructor-1.3.2 → instructor-1.3.3}/instructor/utils.py

@@ -37,6 +37,7 @@ class Response(Protocol):
 
 class Provider(Enum):
     OPENAI = "openai"
+    VERTEXAI = "vertexai"
     ANTHROPIC = "anthropic"
     ANYSCALE = "anyscale"
     TOGETHER = "together"
@@ -67,6 +68,8 @@ def get_provider(base_url: str) -> Provider:
         return Provider.GEMINI
     elif "databricks" in str(base_url):
         return Provider.DATABRICKS
+    elif "vertexai" in str(base_url):
+        return Provider.VERTEXAI
     return Provider.UNKNOWN
 
 
@@ -235,9 +238,13 @@ def transform_to_gemini_prompt(
         if message["role"] == "system":
             system_prompt = message["content"]
         elif message["role"] == "user":
-            messages_gemini.append({"role": "user", "parts": [message["content"]]})
+            messages_gemini.append(
+                {"role": "user", "parts": [message.get("content", "")]}
+            )
         elif message["role"] == "assistant":
-            messages_gemini.append({"role": "model", "parts": [message["content"]]})
+            messages_gemini.append(
+                {"role": "model", "parts": [message.get("content", "")]}
+            )
     if system_prompt:
         messages_gemini[0]["parts"].insert(0, f"*{system_prompt}*")
 

{instructor-1.3.2 → instructor-1.3.3}/pyproject.toml

@@ -1,11 +1,11 @@
 [tool.poetry]
 name = "instructor"
-version = "1.3.2"
+version = "1.3.3"
 description = "structured outputs for llm"
 authors = ["Jason Liu <jason@jxnl.co>"]
 license = "MIT"
 readme = "README.md"
-packages = [{include = "instructor"}]
+packages = [{ include = "instructor" }]
 repository = "https://github.com/jxnl/instructor"
 
 [tool.poetry.dependencies]
@@ -18,6 +18,7 @@ rich = "^13.7.0"
 aiohttp = "^3.9.1"
 tenacity = "^8.2.3"
 pydantic-core = "^2.18.0"
+jiter = "^0.4.1"
 
 # dependency versions for extras
 fastapi = { version = "^0.109.2", optional = true }
@@ -27,21 +28,36 @@ pandas = { version = "^2.2.0", optional = true }
 tabulate = { version = "^0.9.0", optional = true }
 pydantic_extra_types = { version = "^2.6.0", optional = true }
 litellm = { version = "^1.35.31", optional = true }
-anthropic = { version = "^0.26.0", optional = true }
+anthropic = { version = "^0.27.0", optional = true }
 xmltodict = { version = "^0.13.0", optional = true }
 groq = { version = "^0.4.2", optional = true }
 cohere = { version = "^5.1.8", optional = true }
 mistralai = { version = "^0.1.8", optional = true }
 google-generativeai = { version = "^0.5.4", optional = true }
+google-cloud-aiplatform = { version = "^1.52.0", optional = true }
+jsonref = { version = "^1.1.0", optional = true }
 
 [tool.poetry.extras]
 anthropic = ["anthropic", "xmltodict"]
 groq = ["groq"]
 cohere = ["cohere"]
-test-docs = ["fastapi", "redis", "diskcache", "pandas", "tabulate", "pydantic_extra_types", "litellm", "anthropic", "groq", "cohere", "mistralai"]
+test-docs = [
+    "fastapi",
+    "redis",
+    "diskcache",
+    "pandas",
+    "tabulate",
+    "pydantic_extra_types",
+    "litellm",
+    "anthropic",
+    "groq",
+    "cohere",
+    "mistralai",
+]
 mistralai = ["mistralai"]
 litellm = ["litellm"]
 google-generativeai = ["google-generativeai"]
+vertexai = ["google-cloud-aiplatform", "jsonref"]
 
 [tool.poetry.scripts]
 instructor = "instructor.cli.cli:app"
@@ -51,10 +67,11 @@ pytest = "^7.4.0"
 pytest-asyncio = "^0.21.1"
 coverage = "^7.3.2"
 pyright = "^1.1.360"
+jsonref = "^1.1.0"
 
 [tool.poetry.group.docs.dependencies]
 mkdocs = "^1.4.3"
-mkdocs-material = {extras = ["imaging"], version = "^9.5.9"}
+mkdocs-material = { extras = ["imaging"], version = "^9.5.9" }
 mkdocstrings = "^0.22.0"
 mkdocstrings-python = "^1.1.2"
 pytest-examples = "^0.0.10"
@@ -64,7 +81,7 @@ mkdocs-minify-plugin = "^0.8.0"
 mkdocs-redirects = "^1.2.1"
 
 [tool.poetry.group.anthropic.dependencies]
-anthropic = "^0.26.0"
+anthropic = "^0.27.0"
 
 [tool.poetry.group.test-docs.dependencies]
 fastapi = "^0.109.2"
@@ -74,7 +91,7 @@ pandas = "^2.2.0"
 tabulate = "^0.9.0"
 pydantic_extra_types = "^2.6.0"
 litellm = "^1.35.31"
-anthropic = "^0.26.0"
+anthropic = "^0.27.0"
 xmltodict = "^0.13.0"
 groq = "^0.4.2"
 phonenumbers = "^8.13.33"
@@ -84,6 +101,12 @@ mistralai = "^0.1.8"
 [tool.poetry.group.litellm.dependencies]
 litellm = "^1.35.31"
 
+[tool.poetry.group.google-generativeai.dependencies]
+google-generativeai = "^0.5.4"
+
+[tool.poetry.group.vertexai.dependencies]
+google-cloud-aiplatform = "^1.52.0"
+
 [build-system]
 requires = ["poetry-core"]
 build-backend = "poetry.core.masonry.api"

instructor-1.3.2/instructor/client_gemini.py

@@ -1,31 +0,0 @@
-from __future__ import annotations
-
-import google.generativeai as genai
-import instructor
-
-from typing import Any
-
-
-def from_gemini(
-    client: genai.GenerativeModel,
-    mode: instructor.Mode = instructor.Mode.GEMINI_JSON,
-    **kwargs: Any,
-) -> instructor.Instructor:
-    assert (
-        mode == instructor.Mode.GEMINI_JSON
-    ), "Mode must be instructor.Mode.GEMINI_JSON"
-
-    assert isinstance(
-        client,
-        (genai.GenerativeModel),
-    ), "Client must be an instance of genai.generativemodel"
-
-    create = client.generate_content
-
-    return instructor.Instructor(
-        client=client,
-        create=instructor.patch(create=create, mode=mode),
-        provider=instructor.Provider.GEMINI,
-        mode=mode,
-        **kwargs,
-    )

instructor-1.3.2/instructor/exceptions.py

@@ -1,13 +0,0 @@
-class IncompleteOutputException(Exception):
-    """Exception raised when the output from LLM is incomplete due to max tokens limit reached."""
-
-    def __init__(
-        self,
-        *args,
-        last_completion=None,
-        message: str = "The output is incomplete due to a max_tokens length limit.",
-        **kwargs,
-    ):
-        self.last_completion = last_completion
-        super().__init__(message, *args, **kwargs)
-        super().__init__(*args, **kwargs)