model.py

Table of Contents

1. Introduction
2. "import" statements
3. Functions
4. Checklist of features across Streamlit and LocalWorkflow

1. Introduction

• Implements LangChain
• for implementing example selection, prompting, LLMChain, and more.
• Reason - This makes the development capable of expanding on a FOSS framework which increases mutability (such as between choice of LLMs)

2. "import" statements

[To be completed]

3. Functions and classes

Global variables

• hf_global | global scope | used to store HuggingFacePipeline object
• llm_global | global scope | stores choice of llm - either OpenAI or

CustomExampleSelector

• class created on the basis of LangChain documentation for creating a kind of self-querying retriever
• takes the format chosen as the "input_variable" and depending on that returns the examples to be used in the RAG Chain

instantiate_model

• reads the value in llm_global
• creates a huggingface object for the transformer pipeline (in case of LocalWorkflow only) or returns an OpenAI chain

send_request

• forms the interface for passing the prompts to the chat models for output
• constructs a chat prompt template using the format chosen and by creating an object of the CustomExampleSelector class
• passes the chat prompt template to an LLMChain
• Reason - can be used to create more complex RAG Chains in the future

RAGPrompt

• Runs a system prompt which contains the definition as well as extract of a metamodel from the LegalRuleML documentation
• Filters through a list of metamodel definition and XML through a custom Retriever object which is a self-querying retriever made using LangChain's example for CustomRetriever class
• input_dict is a dictionary that contains the XML for alignment with the metamodel.

4. Checklist of features across formats

(TBD = To Be Developed)

Feature	Streamlit	LocalWorkflow	LegalDocML	LegalRuleML
OpenAI option	Yes	Yes	Yes	Yes
HuggingFaceOption	No	Yes	Yes	Yes
Metamodel	Yes	Yes	No	Yes
Example selection	Yes	Yes	TBD	TBD