HumanIQ

This repository is dedicated to a generalized approach of Human-IQ, where users can define their own Langchain pipeline with the flexibility to choose from multiple methods of few-shot examples, including Determinantal Point Processes (DPP) and sentence similarity. The goal of this project is to provide a customizable and extensible framework for building Langchain-based applications.

File Structure

The project consists of the following files:

base_agent.py: This file contains the abstract base class BaseAgent that defines the interface for the agent. It includes abstract methods for planning and tool run logging.
custom_agent.py: This file contains the CustomAgent class, which is a concrete implementation of BaseAgent. It defines the behavior of the agent, including input handling, planning, and execution.
output_parsers.py: This file contains the abstract base class BaseOutputParser and its concrete implementation CustomOutputParser. These classes are responsible for parsing the output of the language model and extracting relevant information.
prompt_templates.py: This file contains the CustomPromptTemplate class, which is used to generate prompts for the language model based on the provided template and tools.
template_construction.py: This file contains the abstract base class BaseTemplateConstruction and its concrete implementation TemplateConstruction. These classes are responsible for constructing the prompt templates used by the agent.
langchain_impl.py: This file contains the abstract base class BaseLangchainImpl and its concrete implementation LangchainImpl. These classes provide the high-level interface for using the Langchain library and executing the agent.
tool_provider.py: This file contains the abstract base class BaseToolProvider and its concrete implementation CustomToolProvider. These classes allow users to define their own tools and provide them to the agent.

Usage

To use the Human-IQ framework, follow these steps:

Clone the repository:

git clone https://github.com/NIMI-research/Human-IQ.git

Install the required dependencies:
```
pip install -r requirements.txt
```

Define your custom tools by creating instances of Tool and providing them to the CustomToolProvider:

from langchain.tools import Tool
from tool_provider import CustomToolProvider

def  wikipedia_search(query):
    # Custom wikipedia search implementation
    pass

def get_wikidata_id(string):
    # Custom implementation
    pass

custom_tools = [
    Tool(
        name="wikipedia_search",
        func=wikipedia_search,
        description="Useful for performing custom searches over wikipedia.",
    ),
    Tool(
        name="get_wikidata_id",
        func=get_wikidata_id,
        description="Useful to retrieve wikidata entity ID given their label.",
    ),
]

tool_provider = CustomToolProvider(custom_tools)

Create an instance of LangchainImpl with the desired parameters and execute the pipeline:

from langchain_impl import LangchainImpl

dataset = "mintaka"
model_name = "gpt-4"
dynamic = True
DPP = False

langchain_impl = LangchainImpl(dataset, model_name, tool_provider, dynamic, DPP)
question = "Who was Kentucky's first governor?"
out, answer_template_for_inference, completion_tokens = langchain_impl.execute_agent(question)

The HumanIQ framework provides several options for customization:

Custom Tools: Define your own tools by creating instances of Tool and providing them to the CustomToolProvider. This allows you to extend the functionality of the Langchain pipeline based on your specific use case.
Few-Shot Examples: Choose from multiple methods of few-shot examples, including Determinantal Point Processes (DPP) and sentence similarity. Modify the TemplateConstruction class in template_construction.py to implement your desired method.
Custom Agents: Implement your own custom agents by extending the BaseAgent class in base_agent.py and providing your implementation in a new file.
Custom Output Parsers: Implement your own output parsers by extending the BaseOutputParser class in output_parsers.py and providing your implementation in a new file.