instructor 1.2.2__tar.gz → 1.2.4__tar.gz

{instructor-1.2.2 → instructor-1.2.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: instructor
-Version: 1.2.2
+Version: 1.2.4
 Summary: structured outputs for llm
 Home-page: https://github.com/jxnl/instructor
 License: MIT
@@ -54,9 +54,9 @@ Instructor is a Python library that makes it a breeze to work with structured ou
 
 - **Response Models**: Specify Pydantic models to define the structure of your LLM outputs
 - **Retry Management**: Easily configure the number of retry attempts for your requests
-   **Validation**: Ensure LLM responses conform to your expectations with Pydantic validation
+- **Validation**: Ensure LLM responses conform to your expectations with Pydantic validation
 - **Streaming Support**: Work with Lists and Partial responses effortlessly
-   **Flexible Backends**: Seamlessly integrate with various LLM providers beyond OpenAI
+- **Flexible Backends**: Seamlessly integrate with various LLM providers beyond OpenAI
 
 ## Get Started in Minutes
 
@@ -201,7 +201,7 @@ assert resp.name == "Jason"
 assert resp.age == 25
 ```
 
-## Type are infered correctly
+## Type are inferred correctly
 
 This was the dream of instructor but due to the patching of openai, it wasnt possible for me to get typing to work well. Now, with the new client, we can get typing to work well! We've also added a few `create_*` methods to make it easier to create iterables and partials, and to access the original completion.
 
@@ -229,7 +229,7 @@ user = client.chat.completions.create(
 )
 ```
 
-Now if you use a IDE, you can see the type is correctly infered.
+Now if you use a IDE, you can see the type is correctly inferred.
 
 ![type](./docs/blog/posts/img/type.png)
 
@@ -341,7 +341,7 @@ for user in user_stream:
     # name='John Doe' age=30
 ```
 
-Notice now that the type infered is `Generator[User, None]`
+Notice now that the type inferred is `Generator[User, None]`
 
 ![generator](./docs/blog/posts/img/generator.png)
 

{instructor-1.2.2 → instructor-1.2.4}/README.md

@@ -11,9 +11,9 @@ Instructor is a Python library that makes it a breeze to work with structured ou
 
 - **Response Models**: Specify Pydantic models to define the structure of your LLM outputs
 - **Retry Management**: Easily configure the number of retry attempts for your requests
-   **Validation**: Ensure LLM responses conform to your expectations with Pydantic validation
+- **Validation**: Ensure LLM responses conform to your expectations with Pydantic validation
 - **Streaming Support**: Work with Lists and Partial responses effortlessly
-   **Flexible Backends**: Seamlessly integrate with various LLM providers beyond OpenAI
+- **Flexible Backends**: Seamlessly integrate with various LLM providers beyond OpenAI
 
 ## Get Started in Minutes
 
@@ -158,7 +158,7 @@ assert resp.name == "Jason"
 assert resp.age == 25
 ```
 
-## Type are infered correctly
+## Type are inferred correctly
 
 This was the dream of instructor but due to the patching of openai, it wasnt possible for me to get typing to work well. Now, with the new client, we can get typing to work well! We've also added a few `create_*` methods to make it easier to create iterables and partials, and to access the original completion.
 
@@ -186,7 +186,7 @@ user = client.chat.completions.create(
 )
 ```
 
-Now if you use a IDE, you can see the type is correctly infered.
+Now if you use a IDE, you can see the type is correctly inferred.
 
 ![type](./docs/blog/posts/img/type.png)
 
@@ -298,7 +298,7 @@ for user in user_stream:
     # name='John Doe' age=30
 ```
 
-Notice now that the type infered is `Generator[User, None]`
+Notice now that the type inferred is `Generator[User, None]`
 
 ![generator](./docs/blog/posts/img/generator.png)
 

{instructor-1.2.2 → instructor-1.2.4}/instructor/client.py

@@ -67,6 +67,7 @@ class Instructor:
         messages: List[ChatCompletionMessageParam],
         max_retries: int = 3,
         validation_context: dict | None = None,
+        strict: bool = True,
         **kwargs,
     ) -> T:
         kwargs = self.handle_kwargs(kwargs)
@@ -76,6 +77,7 @@ class Instructor:
             messages=messages,
             max_retries=max_retries,
             validation_context=validation_context,
+            strict=strict,
             **kwargs,
         )
 
@@ -85,6 +87,7 @@ class Instructor:
         messages: List[ChatCompletionMessageParam],
         max_retries: int = 3,
         validation_context: dict | None = None,
+        strict: bool = True,
         **kwargs,
     ) -> Generator[T, None, None]:
         assert self.provider != Provider.ANTHROPIC, "Anthropic doesn't support partial"
@@ -99,6 +102,7 @@ class Instructor:
             response_model=response_model,
             max_retries=max_retries,
             validation_context=validation_context,
+            strict=strict,
             **kwargs,
         )
 
@@ -108,6 +112,7 @@ class Instructor:
         response_model: Type[T],
         max_retries: int = 3,
         validation_context: dict | None = None,
+        strict: bool = True,
         **kwargs,
     ) -> Iterable[T]:
         assert self.provider != Provider.ANTHROPIC, "Anthropic doesn't support iterable"
@@ -121,6 +126,7 @@ class Instructor:
             response_model=response_model,
             max_retries=max_retries,
             validation_context=validation_context,
+            strict=strict,
             **kwargs,
         )
 
@@ -130,6 +136,7 @@ class Instructor:
         response_model: Type[T],
         max_retries: int = 3,
         validation_context: dict | None = None,
+        strict: bool = True,
         **kwargs,
     ) -> Tuple[T, ChatCompletion | Any]:
         kwargs = self.handle_kwargs(kwargs)
@@ -138,6 +145,7 @@ class Instructor:
             response_model=response_model,
             max_retries=max_retries,
             validation_context=validation_context,
+            strict=strict,
             **kwargs,
         )
         return model, model._raw_response
@@ -176,6 +184,7 @@ class AsyncInstructor(Instructor):
         response_model: Type[T],
         validation_context: dict | None = None,
         max_retries: int = 3,
+        strict: bool = True,
         **kwargs,
     ) -> T:
         kwargs = self.handle_kwargs(kwargs)
@@ -184,6 +193,7 @@ class AsyncInstructor(Instructor):
             validation_context=validation_context,
             max_retries=max_retries,
             messages=messages,
+            strict=strict,
             **kwargs,
         )
 
@@ -193,6 +203,7 @@ class AsyncInstructor(Instructor):
         messages: List[ChatCompletionMessageParam],
         validation_context: dict | None = None,
         max_retries: int = 3,
+        strict: bool = True,
         **kwargs,
     ) -> AsyncGenerator[T, None]:
         assert self.provider != Provider.ANTHROPIC, "Anthropic doesn't support partial"
@@ -204,6 +215,7 @@ class AsyncInstructor(Instructor):
             validation_context=validation_context,
             max_retries=max_retries,
             messages=messages,
+            strict=strict,
             **kwargs,
         ):
             yield item
@@ -214,6 +226,7 @@ class AsyncInstructor(Instructor):
         messages: List[ChatCompletionMessageParam],
         validation_context: dict | None = None,
         max_retries: int = 3,
+        strict: bool = True,
         **kwargs,
     ) -> AsyncGenerator[T, None]:
         assert self.provider != Provider.ANTHROPIC, "Anthropic doesn't support iterable"
@@ -225,6 +238,7 @@ class AsyncInstructor(Instructor):
             validation_context=validation_context,
             max_retries=max_retries,
             messages=messages,
+            strict=strict,
             **kwargs,
         ):
             yield item
@@ -235,6 +249,7 @@ class AsyncInstructor(Instructor):
         messages: List[ChatCompletionMessageParam],
         validation_context: dict | None = None,
         max_retries: int = 3,
+        strict: bool = True,
         **kwargs,
     ) -> Tuple[T, dict]:
         kwargs = self.handle_kwargs(kwargs)
@@ -243,6 +258,7 @@ class AsyncInstructor(Instructor):
             validation_context=validation_context,
             max_retries=max_retries,
             messages=messages,
+            strict=strict,
             **kwargs,
         )
         return response, response._raw_response
@@ -269,11 +285,18 @@ def from_openai(
     mode: instructor.Mode = instructor.Mode.TOOLS,
     **kwargs,
 ) -> Instructor | AsyncInstructor:
-    provider = get_provider(str(client.base_url))
+    if hasattr(client, "base_url"):
+        provider = get_provider(str(client.base_url))
+    else:
+        provider = Provider.OPENAI
+
+    if not isinstance(client, (openai.OpenAI, openai.AsyncOpenAI)):
+        import warnings
 
-    assert isinstance(
-        client, (openai.OpenAI, openai.AsyncOpenAI)
-    ), "Client must be an instance of openai.OpenAI or openai.AsyncOpenAI"
+        warnings.warn(
+            "Client should be an instance of openai.OpenAI or openai.AsyncOpenAI. "
+            "Unexpected behavior may occur with other client types."
+        )
 
     if provider in {Provider.ANYSCALE, Provider.TOGETHER}:
         assert mode in {
@@ -316,8 +339,7 @@ def from_litellm(
     completion: Callable,
     mode: instructor.Mode = instructor.Mode.TOOLS,
     **kwargs,
-) -> Instructor:
-    ...
+) -> Instructor: ...
 
 
 @overload

{instructor-1.2.2 → instructor-1.2.4}/instructor/function_calls.py

@@ -1,15 +1,14 @@
-from typing import Any, Dict, Optional, Type, TypeVar
-from docstring_parser import parse
+import logging
 from functools import wraps
-from pydantic import BaseModel, create_model
+from typing import Annotated, Any, Dict, Optional, Type, TypeVar
+
+from docstring_parser import parse
 from openai.types.chat import ChatCompletion
-from typing import Any, Dict, Optional, Type
-from instructor.mode import Mode
-from instructor.utils import extract_json_from_codeblock
+from pydantic import BaseModel, Field, TypeAdapter, create_model
+
 from instructor.exceptions import IncompleteOutputException
 from instructor.mode import Mode
-import logging
-
+from instructor.utils import extract_json_from_codeblock
 
 T = TypeVar("T")
 
@@ -119,7 +118,12 @@ class OpenAISchema(BaseModel):  # type: ignore[misc]
         validation_context: Optional[Dict[str, Any]] = None,
         strict: Optional[bool] = None,
     ) -> BaseModel:
-        tool_call = [c.input for c in completion.content if c.type == "tool_use"][0]
+        tool_calls = [c.input for c in completion.content if c.type == "tool_use"]
+
+        tool_calls_validator = TypeAdapter(
+            Annotated[list, Field(min_length=1, max_length=1)]
+        )
+        tool_call = tool_calls_validator.validate_python(tool_calls)[0]
 
         return cls.model_validate(tool_call, context=validation_context, strict=strict)  # type:ignore
 

{instructor-1.2.2 → instructor-1.2.4}/instructor/patch.py

@@ -116,6 +116,7 @@ def patch(
         response_model: Type[T_Model] = None,
         validation_context: dict = None,
         max_retries: int = 1,
+        strict: bool = True,
         *args: T_ParamSpec.args,
         **kwargs: T_ParamSpec.kwargs,
     ) -> T_Model:
@@ -129,6 +130,7 @@ def patch(
             max_retries=max_retries,
             args=args,
             kwargs=new_kwargs,
+            strict=strict,
             mode=mode,
         )  # type: ignore
         return response
@@ -138,6 +140,7 @@ def patch(
         response_model: Type[T_Model] = None,
         validation_context: dict = None,
         max_retries: int = 1,
+        strict: bool = True,
         *args: T_ParamSpec.args,
         **kwargs: T_ParamSpec.kwargs,
     ) -> T_Model:
@@ -150,6 +153,7 @@ def patch(
             validation_context=validation_context,
             max_retries=max_retries,
             args=args,
+            strict=strict,
             kwargs=new_kwargs,
             mode=mode,
         )

{instructor-1.2.2 → instructor-1.2.4}/instructor/retry.py

@@ -62,22 +62,27 @@ def reask_messages(response: ChatCompletion, mode: Mode, exception: Exception):
             ):
                 tool_use_id = content.id
 
-        assert tool_use_id is not None, "Tool use ID not found in the response"
         yield {
             "role": "assistant",
             "content": assistant_content,
         }
-        yield {
-            "role": "user",
-            "content": [
-                {
-                    "type": "tool_result",
-                    "tool_use_id": tool_use_id,
-                    "content": f"Validation Error found:\n{exception}\nRecall the function correctly, fix the errors",
-                    "is_error": True,
-                }
-            ],
-        }
+        if tool_use_id is not None:
+            yield {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "tool_result",
+                        "tool_use_id": tool_use_id,
+                        "content": f"Validation Error found:\n{exception}\nRecall the function correctly, fix the errors",
+                        "is_error": True,
+                    }
+                ],
+            }
+        else:
+            yield {
+                "role": "user",
+                "content": f"Validation Error due to no tool invocation:\n{exception}\nRecall the function correctly, fix the errors",
+            }
         return
     if mode == Mode.ANTHROPIC_JSON:
         from anthropic.types import Message
@@ -128,6 +133,10 @@ def retry_sync(
     mode: Mode = Mode.TOOLS,
 ) -> T_Model:
     total_usage = CompletionUsage(completion_tokens=0, prompt_tokens=0, total_tokens=0)
+    if mode in {Mode.ANTHROPIC_TOOLS, Mode.ANTHROPIC_JSON}:
+        from anthropic.types import Usage as AnthropicUsage
+
+        total_usage = AnthropicUsage(input_tokens=0, output_tokens=0)
 
     # If max_retries is int, then create a Retrying object
     if isinstance(max_retries, int):
@@ -189,6 +198,11 @@ async def retry_async(
     mode: Mode = Mode.TOOLS,
 ) -> T:
     total_usage = CompletionUsage(completion_tokens=0, prompt_tokens=0, total_tokens=0)
+    if mode in {Mode.ANTHROPIC_TOOLS, Mode.ANTHROPIC_JSON}:
+        from anthropic.types import Usage as AnthropicUsage
+
+        total_usage = AnthropicUsage(input_tokens=0, output_tokens=0)
+
     # If max_retries is int, then create a AsyncRetrying object
     if isinstance(max_retries, int):
         logger.debug(f"max_retries: {max_retries}")
@@ -227,7 +241,7 @@ async def retry_async(
                     raise InstructorRetryException(
                         e,
                         last_completion=response,
-                        n_attempts=e.attempt_number,
+                        n_attempts=attempt.retry_state.attempt_number,
                         messages=kwargs["messages"],
                         total_usage=total_usage,
                     ) from e
@@ -236,7 +250,7 @@ async def retry_async(
         raise InstructorRetryException(
             e,
             last_completion=response,
-            n_attempts=e.attempt_number,
+            n_attempts=attempt.retry_state.attempt_number,
             messages=kwargs["messages"],
             total_usage=total_usage,
         ) from e

{instructor-1.2.2 → instructor-1.2.4}/instructor/utils.py

@@ -2,16 +2,18 @@ from __future__ import annotations
 
 import inspect
 import json
+import logging
 from typing import Callable, Generator, Iterable, AsyncGenerator, TypeVar
 
 from pydantic import BaseModel
-
+from openai.types import CompletionUsage as OpenAIUsage
 from openai.types.chat import (
     ChatCompletion,
     ChatCompletionMessage,
     ChatCompletionMessageParam,
 )
 
+logger = logging.getLogger("instructor")
 T_Model = TypeVar("T_Model", bound=BaseModel)
 
 from enum import Enum
@@ -93,11 +95,27 @@ async def extract_json_from_stream_async(
 
 
 def update_total_usage(response: T_Model, total_usage) -> T_Model | ChatCompletion:
-    if isinstance(response, ChatCompletion) and response.usage is not None:
-        total_usage.completion_tokens += response.usage.completion_tokens or 0
-        total_usage.prompt_tokens += response.usage.prompt_tokens or 0
-        total_usage.total_tokens += response.usage.total_tokens or 0
+    response_usage = getattr(response, "usage", None)
+    if isinstance(response_usage, OpenAIUsage):
+        total_usage.completion_tokens += response_usage.completion_tokens or 0
+        total_usage.prompt_tokens += response_usage.prompt_tokens or 0
+        total_usage.total_tokens += response_usage.total_tokens or 0
         response.usage = total_usage  # Replace each response usage with the total usage
+        return response
+
+    # Anthropic usage
+    try:
+        from anthropic.types import Usage as AnthropicUsage
+
+        if isinstance(response_usage, AnthropicUsage):
+            total_usage.input_tokens += response_usage.input_tokens or 0
+            total_usage.output_tokens += response_usage.output_tokens or 0
+            response.usage = total_usage
+            return response
+    except ImportError:
+        pass
+
+    logger.debug("No compatible response.usage found, token usage not updated.")
     return response
 
 

{instructor-1.2.2 → instructor-1.2.4}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "instructor"
-version = "1.2.2"
+version = "1.2.4"
 description = "structured outputs for llm"
 authors = ["Jason Liu <jason@jxnl.co>"]
 license = "MIT"