agentics-py 0.0.1__py3-none-any.whl

agentics/__init__.py

@@ -0,0 +1 @@
+from .core import *

agentics/abstractions/pydantic_transducer.py

@@ -0,0 +1,221 @@
+import asyncio
+import json
+import os
+from collections.abc import Iterable
+from typing import Any, Optional, Type, Union
+
+from crewai import Agent, Crew, Process, Task
+from dotenv import load_dotenv
+from loguru import logger
+from openai import AsyncOpenAI
+from pydantic import BaseModel
+
+from agentics.core.llm_connections import watsonx_crewai_llm
+from agentics.core.utils import openai_response
+
+load_dotenv()
+
+
+class PydanticTransducerVLLM:
+    model_config = {"arbitrary_types_allowed": True}
+    llm: AsyncOpenAI
+    intensional_definiton: str
+    verbose: bool = False
+    MAX_CHAR_PROMPT: int = 15000
+
+    def __init__(
+        self,
+        atype: Type[BaseModel],
+        verbose: bool = False,
+        llm=None,
+        tools=None,
+        intensional_definiton=None,
+        **kwargs,
+    ):
+        self.atype = atype
+        self.verbose = verbose
+        self.llm = llm
+        self.tools = tools
+        self.intensional_definiton = (
+            intensional_definiton
+            if intensional_definiton
+            else """Generate an object of the specified Pydantic Type from the following input."""
+        )
+        self.llm_params = {
+            "extra_body": {"guided_json": self.atype.model_json_schema()},
+            "logprobs": False,
+            "n": 1,
+        }
+        self.llm_params.update(kwargs)
+
+    async def async_transduce(
+        self,
+        input: Union[str, list[Any]],
+        logprobs: bool = False,
+        n_samples: int = 1,
+        **kwargs,
+    ) -> list[list[BaseModel]]:
+
+        default_user_prompt = "\n".join(
+            [
+                self.intensional_definiton,
+                "Generate an object of the specified Pydantic Type from the following input.\n",
+            ]
+        )
+        self.llm_params.update(kwargs)
+        if isinstance(input, str):
+            result = await openai_response(
+                model=os.getenv("VLLM_MODEL_ID"),
+                base_url=os.getenv("VLLM_URL"),
+                user_prompt=default_user_prompt + str(state),
+                **self.llm_params,
+            )
+            decoded_result = self.atype.model_validate_json(result)
+            return decoded_result
+
+        elif isinstance(input, Iterable) and all(isinstance(i, str) for i in input):
+            processes = []
+            for state in input:
+                corutine = openai_response(
+                    model=os.getenv("VLLM_MODEL_ID"),
+                    base_url=os.getenv("VLLM_URL"),
+                    user_prompt=default_user_prompt + str(state),
+                    **self.llm_params,
+                )
+                processes.append(corutine)
+            results = await asyncio.wait_for(
+                asyncio.gather(*processes, return_exceptions=True), timeout=10000
+            )
+
+            decoded_results = []
+            for result in results:
+                if issubclass(type(result), Exception):
+                    if self.verbose:
+                        logger.debug("Something went wrongs, generating empty states")
+                    decoded_results.append(self.atype())
+                else:
+                    decoded_results.append(self.atype.model_validate_json(result))
+            return decoded_results
+        else:
+            return NotImplemented
+
+
+class PydanticTransducerCrewAI:
+    crew: Crew
+    llm: Any
+    intensional_definiton: str
+    verbose: bool = False
+    MAX_CHAR_PROMPT: int = 15000
+
+    def __init__(
+        self,
+        atype: Type[BaseModel],
+        verbose: bool = False,
+        llm=None,
+        tools=None,
+        intensional_definiton=None,
+        **kwargs,
+    ):
+        self.atype = atype
+        self.llm = llm if llm else watsonx_crewai_llm
+        self.intensional_definiton = (
+            intensional_definiton
+            if intensional_definiton
+            else """Generate an object of the specified Pydantic Type from the following input."""
+        )
+        self.prompt_params = {
+            "role": "Task Executor",
+            "goal": "You execute tasks",
+            "backstory": "You are always faithful and provide only fact based answers.",
+            "expected_output": "Described by Pydantic Type",
+        }
+        self.prompt_params.update(kwargs)
+        agent = Agent(
+            role=self.prompt_params["role"],
+            goal=self.prompt_params["goal"],
+            backstory=self.prompt_params["backstory"],
+            verbose=verbose,
+            max_iter=3,
+            llm=self.llm,
+            tools=tools if tools else [],
+        )
+        task = Task(
+            description=self.intensional_definiton + " {task_description}",
+            expected_output=self.prompt_params["expected_output"],
+            output_file="",
+            agent=agent,
+            output_pydantic=self.atype,
+            tools=tools,
+        )
+        self.crew = Crew(
+            agents=[agent],
+            tasks=[task],
+            process=Process.sequential,
+            verbose=verbose,
+            manager_llm=self.llm,
+            function_calling_llm=self.llm,
+            chat_llm=self.llm,
+        )
+
+    async def __kickoff_with_index(self, state, i):
+        return (await self.crew.kickoff_async(state), i, state)
+
+    async def async_transduce(self, input):
+        if isinstance(input, str):
+            answer = self.crew.kickoff_async({"task_description": input})
+            ans = await answer
+            return ans.pydantic
+        elif isinstance(input, Iterable) and all(isinstance(i, str) for i in input):
+            input_states = [
+                {"task_description": x[: self.MAX_CHAR_PROMPT]} for x in input
+            ]
+            answer_list = await self.crew.kickoff_for_each_async(input_states)
+
+            return [x.pydantic for x in answer_list]
+        else:
+            return NotImplemented
+
+    def transduce(self, input):
+        answer = self.crew.kickoff({"task_description": input})
+        return answer.pydantic
+
+
+import asyncio
+
+if __name__ == "__main__":
+
+    class PersonalInformation(BaseModel):
+        first_name: Optional[str] = None
+        last_name: Optional[str] = None
+        year_of_birth: Optional[int] = None
+        nationality: Optional[str] = None
+
+    pt = PydanticTransducerCrewAI(PersonalInformation, llm=None)
+    print(
+        asyncio.run(
+            pt.async_transduce(
+                [
+                    """Hi , I am John. My father is Dave Smith. 
+                                I was born in April 1977, in British Columbia """,
+                    """Hi , I am John. My father is Dave Smith. 
+                                I was born in April 1977, in British Columbia """,
+                ]
+                * 5
+            )
+        )
+    )
+    # from agentics.core.llm_connections import vllm_llm
+    # pt = PydanticTransducerVLLM(atype=PersonalInformation, llm=vllm_llm)
+    # print(
+    #     asyncio.run(
+    #         pt.async_transduce(
+    #             [
+    #                 """Hi , I am John. My father is Dave Smith.
+    #                                I was born in April 1977, in British Columbia """,
+    #                 """Hi , I am John. My father is Dave Smith.
+    #                                I was born in April 1977, in British Columbia """,
+    #             ]
+    #             * 10
+    #         )
+    #     )
+    # )

agentics/abstractions/structured_output.py

@@ -0,0 +1,13 @@
+from typing import Type
+
+from crewai import LLM
+from pydantic import BaseModel
+
+from agentics.core.llm_connections import watsonx_crewai_llm
+
+
+async def generate_structured_output(
+    prompt: str, output_type: Type[BaseModel], llm: LLM = watsonx_crewai_llm
+):
+    structured_llm = llm.with_structured_output(output_type)
+    return await structured_llm.ainvoke(prompt)

agentics/core/__init__.py

@@ -0,0 +1 @@
+from .agentics import Agentics