build json & pdf reader robustness && ruff

concepts/textbook-generator.py

@@ -8,40 +8,40 @@
 
 # This example will individually import all the necessary modules from zyx
 from zyx import (
-    scrape, # For generating our web research
-    BaseModel, Field # For generating our textbook & model definitions
+    scrape,  # For generating our web research
+    BaseModel,
+    Field,  # For generating our textbook & model definitions
 )
 
 
 def generate_textbook(
-        instructions : str,
-        model : str = "gpt-4o-mini",
+    instructions: str,
+    model: str = "gpt-4o-mini",
 ) -> BaseModel:
-
     # Lets first collect our information
     ai_research = scrape(
         instructions,
-        model = "gpt-4o-mini",
-        num_queries = 5, # This determines how many queries are generated based on our original query
-        workers = 5, # This determines how many threads are used for our web search
-        max_results = 5, # This determines the maximum number of results we scrape from the web
-        verbose = True # This determines if we want to print out the raw results as we scrape them
+        model="gpt-4o-mini",
+        num_queries=5,  # This determines how many queries are generated based on our original query
+        workers=5,  # This determines how many threads are used for our web search
+        max_results=5,  # This determines the maximum number of results we scrape from the web
+        verbose=True,  # This determines if we want to print out the raw results as we scrape them
     )
 
     # Now that we have our research; lets generate our textbook
     # First, we'll have to define our textbook using Pydantic
     class TextbookChapter(BaseModel):
-        title : str = Field(description="The title of the chapter")
-        content : str = Field(description="The content of the chapter")
-
+        title: str = Field(description="The title of the chapter")
+        content: str = Field(description="The content of the chapter")
 
     class Textbook(BaseModel):
-        title : str = Field(description="The title of the textbook")
-        chapters : list[TextbookChapter] = Field(description="The chapters of the textbook")
-
+        title: str = Field(description="The title of the textbook")
+        chapters: list[TextbookChapter] = Field(
+            description="The chapters of the textbook"
+        )
 
     textbook = Textbook.generate(
-        instructions = f"""
+        instructions=f"""
         We have collecting the following information.
 
         <research>
@@ -50,8 +50,8 @@ class Textbook(BaseModel):
 
         Please use this information to create a comprehensive textbook on {instructions}.
         """,
-        model = "gpt-4o-mini",
-        process = "sequential" # This generates each field in the model one by one
+        model="gpt-4o-mini",
+        process="sequential",  # This generates each field in the model one by one
     )
 
     # Lets write our textbook to a markdown file
@@ -63,10 +63,8 @@ class Textbook(BaseModel):
 
 
 if __name__ == "__main__":
-
     textbook = generate_textbook(
-        instructions = "AI & LLMs in September 2024",
-        model = "gpt-4o-mini"
+        instructions="AI & LLMs in September 2024", model="gpt-4o-mini"
     )
 
-    print(textbook)
+    print(textbook)

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "zyx"
-version = "1.0.10"
+version = "1.0.11"
 description = "A hyper-fast, fun, quality-of-life focused & genuinely useful LLM toolkit. Inspired by Marvin-AI. Built on LiteLLM, Instructor & Qdrant."
 authors = ["Hammad Saeed <hvmmad@gmail.com>"]
 readme = "readme.md"

zyx/client.py

@@ -1,5 +1,6 @@
 from .lib.utils.logger import get_logger
 from .lib.utils.convert_to_openai_tool import convert_to_openai_tool
+from instructor import Mode
 
 logger = get_logger(__name__)
 
@@ -22,31 +23,6 @@
 ## -- Instructor Configuration -- ##
 
 
-## -- Instructor Mode -- ##
-## This was directly ported from instructor
-## https://github.com/jxnl/instructor/
-class Mode(enum.Enum):
-    """The mode to use for patching the client"""
-
-    FUNCTIONS = "function_call"
-    PARALLEL_TOOLS = "parallel_tool_call"
-    TOOLS = "tool_call"
-    MISTRAL_TOOLS = "mistral_tools"
-    JSON = "json_mode"
-    JSON_O1 = "json_o1"
-    MD_JSON = "markdown_json_mode"
-    JSON_SCHEMA = "json_schema_mode"
-    ANTHROPIC_TOOLS = "anthropic_tools"
-    ANTHROPIC_JSON = "anthropic_json"
-    COHERE_TOOLS = "cohere_tools"
-    VERTEXAI_TOOLS = "vertexai_tools"
-    VERTEXAI_JSON = "vertexai_json"
-    GEMINI_JSON = "gemini_json"
-    GEMINI_TOOLS = "gemini_tools"
-    COHERE_JSON_SCHEMA = "json_object"
-    TOOLS_STRICT = "tools_strict"
-
-
 InstructorMode = Literal[
     "function_call",
     "parallel_tool_call",
@@ -487,7 +463,7 @@ def __init_client__(self):
 
         return client
 
-    def __patch_client__(self):
+    def __patch_client__(self, mode: Optional[InstructorMode] = "tool_call"):
         """
         Patches the client with Instructor.
         """
@@ -499,12 +475,16 @@ def __patch_client__(self):
         if self.provider == "openai":
             from instructor import from_openai
 
-            patched_client = from_openai(self.clients.client)
+            patched_client = from_openai(
+                self.clients.client, mode=get_mode(mode) if mode else None
+            )
 
         else:
             from instructor import patch
 
-            patched_client = patch(self.clients.client)
+            patched_client = patch(
+                self.clients.client, mode=get_mode(mode) if mode else None
+            )
 
         if self.config.verbose:
             logger.info(f"Patched {self.provider} client with Instructor")
@@ -594,17 +574,21 @@ def instructor_completion(self, args: CompletionArgs):
             if args.tools is None:
                 exclude_params.update({"tools", "parallel_tool_calls", "tool_choice"})
 
+            if self.config.verbose:
+                logger.info(f"Excluding the following parameters: {exclude_params}")
+                logger.info(f"Args: {args.model_dump(exclude=exclude_params)}")
+
             if args.model.startswith("o1-"):
                 logger.warning(
                     "OpenAI O1- model detected. Removing all non-supported parameters."
                 )
                 exclude_params.update(
                     {
-                    "max_tokens",
-                    "temperature",
-                    "top_p",
-                    "frequency_penalty",
-                    "presence_penalty",
+                        "max_tokens",
+                        "temperature",
+                        "top_p",
+                        "frequency_penalty",
+                        "presence_penalty",
                         "tools",
                         "parallel_tool_calls",
                         "tool_choice",
@@ -804,7 +788,9 @@ def run_completion(
                 args = self.execute_tool_call(formatted_tools, args, base_response)
 
                 if args:
-                    original_args.messages.extend(args.messages[len(original_args.messages):])
+                    original_args.messages.extend(
+                        args.messages[len(original_args.messages) :]
+                    )
                     original_args.response_model = response_model
                     original_args.stream = stream
                     return self.instructor_completion(original_args)
@@ -896,9 +882,6 @@ def completion(
         model = recommended_model
 
         if response_model:
-            if not self.clients.instructor:
-                self.clients.instructor = self.__patch_client__()
-
             mode = get_mode(mode)
 
             if model.startswith("o1-"):
@@ -907,7 +890,14 @@ def completion(
                 )
                 mode = Mode.JSON_O1
 
-            self.clients.instructor.mode = mode
+            if not self.clients.instructor:
+                self.clients.instructor = self.__patch_client__(mode)
+
+            else:
+                self.clients.instructor.mode = mode
+
+            if verbose:
+                logger.info(f"Instructor Mode: {self.clients.instructor.mode}")
 
         return self.run_completion(
             messages=messages,
@@ -1024,6 +1014,9 @@ def _completion(
 
             client.clients.instructor.mode = mode
 
+            if verbose:
+                logger.info(f"Instructor Mode: {mode}")
+
         return client.run_completion(
             messages=messages,
             model=model,

zyx/resources/completions/agents/scrape.py

@@ -12,7 +12,9 @@
 except ImportError:
     import os
 
-    print("The [bold]`zyx(data)`[/bold] data extension is required to use this module. Install it?")
+    print(
+        "The [bold]`zyx(data)`[/bold] data extension is required to use this module. Install it?"
+    )
     if input("Install? (y/n)") == "y":
         os.system("pip install 'zyx[data]'")
     else:

zyx/resources/data/chunk.py

@@ -6,7 +6,9 @@
 except ImportError:
     import os
 
-    print("The [bold]`zyx(data)`[/bold] data extension is required to use this module. Install it?")
+    print(
+        "The [bold]`zyx(data)`[/bold] data extension is required to use this module. Install it?"
+    )
     if input("Install? (y/n)") == "y":
         os.system("pip install 'zyx[data]'")
     else: