instructor 1.3.0__tar.gz → 1.3.2__tar.gz

{instructor-1.3.0 → instructor-1.3.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: instructor
-Version: 1.3.0
+Version: 1.3.2
 Summary: structured outputs for llm
 Home-page: https://github.com/jxnl/instructor
 License: MIT
@@ -244,7 +244,7 @@ assert resp.name == "Jason"
 assert resp.age == 25
 ```
 
-## Type are inferred correctly
+## Types 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.
 
@@ -272,7 +272,7 @@ user = client.chat.completions.create(
 )
 ```
 
-Now if you use a IDE, you can see the type is correctly inferred.
+Now if you use an IDE, you can see the type is correctly inferred.
 
 ![type](./docs/blog/posts/img/type.png)
 
@@ -339,7 +339,7 @@ user, completion = client.chat.completions.create_with_completion(
 
 ### Streaming Partial Objects: `create_partial`
 
-In order to handle streams, we still support `Iterable[T]` and `Partial[T]` but to simply the type inference, we've added `create_iterable` and `create_partial` methods as well!
+In order to handle streams, we still support `Iterable[T]` and `Partial[T]` but to simplify the type inference, we've added `create_iterable` and `create_partial` methods as well!
 
 ```python
 import openai

{instructor-1.3.0 → instructor-1.3.2}/README.md

@@ -198,7 +198,7 @@ assert resp.name == "Jason"
 assert resp.age == 25
 ```
 
-## Type are inferred correctly
+## Types 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.
 
@@ -226,7 +226,7 @@ user = client.chat.completions.create(
 )
 ```
 
-Now if you use a IDE, you can see the type is correctly inferred.
+Now if you use an IDE, you can see the type is correctly inferred.
 
 ![type](./docs/blog/posts/img/type.png)
 
@@ -293,7 +293,7 @@ user, completion = client.chat.completions.create_with_completion(
 
 ### Streaming Partial Objects: `create_partial`
 
-In order to handle streams, we still support `Iterable[T]` and `Partial[T]` but to simply the type inference, we've added `create_iterable` and `create_partial` methods as well!
+In order to handle streams, we still support `Iterable[T]` and `Partial[T]` but to simplify the type inference, we've added `create_iterable` and `create_partial` methods as well!
 
 ```python
 import openai

{instructor-1.3.0 → instructor-1.3.2}/instructor/__init__.py

@@ -52,7 +52,10 @@ if importlib.util.find_spec("anthropic") is not None:
 
     __all__ += ["from_anthropic"]
 
-if importlib.util.find_spec("google.generativeai") is not None:
+if (
+    importlib.util.find_spec("google")
+    and importlib.util.find_spec("google.generativeai") is not None
+):
     from .client_gemini import from_gemini
 
     __all__ += ["from_gemini"]

{instructor-1.3.0 → instructor-1.3.2}/instructor/client_groq.py

@@ -20,7 +20,7 @@ def from_groq(
     client: groq.AsyncGroq,
     mode: instructor.Mode = instructor.Mode.TOOLS,
     **kwargs: Any,
-) -> instructor.Instructor:
+) -> instructor.AsyncInstructor:
     ...
 
 
@@ -28,7 +28,7 @@ def from_groq(
     client: groq.Groq | groq.AsyncGroq,
     mode: instructor.Mode = instructor.Mode.TOOLS,
     **kwargs: Any,
-) -> instructor.Instructor:
+) -> instructor.Instructor | instructor.AsyncInstructor:
     assert mode in {
         instructor.Mode.JSON,
         instructor.Mode.TOOLS,
@@ -48,7 +48,7 @@ def from_groq(
         )
 
     else:
-        return instructor.Instructor(
+        return instructor.AsyncInstructor(
             client=client,
             create=instructor.patch(create=client.chat.completions.create, mode=mode),
             provider=instructor.Provider.GROQ,

{instructor-1.3.0 → instructor-1.3.2}/instructor/process_response.py

@@ -314,9 +314,7 @@ def handle_response_model(
                 + "\n\n".join(openai_system_messages)
             )
 
-            new_kwargs[
-                "system"
-            ] += f"""
+            new_kwargs["system"] += f"""
             You must only response in JSON format that adheres to the following schema:
 
             <JSON_SCHEMA>
@@ -359,6 +357,9 @@ The output must be a valid JSON object that `{response_model.__name__}.model_val
             new_kwargs["message"] = instruction
             new_kwargs["chat_history"] = chat_history
         elif mode == Mode.GEMINI_JSON:
+            assert (
+                "model" not in new_kwargs
+            ), "Gemini `model` must be set while patching the client, not passed as a parameter to the create method"
             message = dedent(
                 f"""
                 As a genius expert, your task is to understand the content and provide
@@ -388,15 +389,34 @@ The output must be a valid JSON object that `{response_model.__name__}.model_val
                 "generation_config", {}
             ) | {"response_mime_type": "application/json"}
 
+            map_openai_args_to_gemini = {
+                "max_tokens": "max_output_tokens",
+                "temperature": "temperature",
+                "n": "candidate_count",
+                "top_p": "top_p",
+                "stop": "stop_sequences",
+            }
+
+            # update gemini config if any params are set
+            for k, v in map_openai_args_to_gemini.items():
+                val = new_kwargs.pop(k, None)
+                if val == None:
+                    continue
+                new_kwargs["generation_config"][v] = val
+
+            # gemini has a different prompt format and params from other providers
+            new_kwargs["contents"] = transform_to_gemini_prompt(
+                new_kwargs.pop("messages")
+            )
+
             # minimize gemini safety related errors - model is highly prone to false alarms
+            from google.generativeai.types import HarmCategory, HarmBlockThreshold
+
             new_kwargs["safety_settings"] = new_kwargs.get("safety_settings", {}) | {
                 HarmCategory.HARM_CATEGORY_HATE_SPEECH: HarmBlockThreshold.BLOCK_ONLY_HIGH,
                 HarmCategory.HARM_CATEGORY_HARASSMENT: HarmBlockThreshold.BLOCK_ONLY_HIGH,
                 HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT: HarmBlockThreshold.BLOCK_ONLY_HIGH,
             }
-            # gemini has a different prompt format and params from other providers
-            new_kwargs["contents"] = transform_to_gemini_prompt(new_kwargs["messages"])
-            del new_kwargs["messages"]
         else:
             raise ValueError(f"Invalid patch mode: {mode}")
 

{instructor-1.3.0 → instructor-1.3.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "instructor"
-version = "1.3.0"
+version = "1.3.2"
 description = "structured outputs for llm"
 authors = ["Jason Liu <jason@jxnl.co>"]
 license = "MIT"