Skip to content

on_boarding.md

Latest commit

 

History

History
102 lines (63 loc) · 7.69 KB

on_boarding.md

File metadata and controls

102 lines (63 loc) · 7.69 KB
graph LR
    LLM_Orchestrator["LLM Orchestrator"]
    Context_Management_Module["Context Management Module"]
    Memory_Management_Module["Memory Management Module"]
    Condition_Evaluation_Module["Condition Evaluation Module"]
    GPT_Selection_Module["GPT Selection Module"]
    Summarization_Module["Summarization Module"]
    LLM_External_["LLM (External)"]
    LLM_Orchestrator -- "Sends Prompts To" --> LLM_External_
    LLM_External_ -- "Returns Responses To" --> LLM_Orchestrator
    Context_Management_Module -- "Prepares Input For" --> LLM_Orchestrator
    LLM_Orchestrator -- "Sends Output To" --> Condition_Evaluation_Module
    LLM_Orchestrator -- "Sends Output To" --> GPT_Selection_Module
    LLM_Orchestrator -- "Stores Output In" --> Memory_Management_Module
    Memory_Management_Module -- "Provides Context To" --> Context_Management_Module
    Context_Management_Module -- "Sends Text For Summarization To" --> Summarization_Module
    Summarization_Module -- "Provides Summarized Text To" --> Context_Management_Module
    LLM_Orchestrator -- "Receives Evaluation Result From" --> Condition_Evaluation_Module
    LLM_Orchestrator -- "Receives Selected Option From" --> GPT_Selection_Module
    click LLM_Orchestrator href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/thinkgpt/LLM_Orchestrator.md" "Details"
    click Memory_Management_Module href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/thinkgpt/Memory_Management_Module.md" "Details"
Loading

CodeBoardingDemoContact

Details

The thinkgpt project is structured around an LLM Orchestrator that serves as the central hub for interacting with external Large Language Models. This orchestrator is supported by several specialized modules: the Context Management Module prepares and optimizes input, integrating information from the Memory Management Module and Summarization Module. The Memory Management Module handles persistent and transient data storage. Post-LLM response, the Condition Evaluation Module and GPT Selection Module interpret and process the output for specific decision-making and conditional logic. The Summarization Module condenses large texts to fit LLM constraints. This modular design ensures efficient, context-aware, and adaptable LLM interactions. The core of the thinkgpt architecture revolves around the LLM Orchestrator, which acts as the primary interface to external LLMs. Data flow is largely initiated by the LLM Orchestrator sending prompts to the LLM (External) and receiving responses. Before prompt submission, the Context Management Module plays a crucial role in preparing the input, drawing upon historical data from the Memory Management Module and condensed information from the Summarization Module. After receiving responses from the LLM, the LLM Orchestrator dispatches these outputs to specialized modules like the Condition Evaluation Module for logical assessment and the GPT Selection Module for decision-making, receiving processed results back. This clear separation of concerns facilitates maintainable and scalable LLM application development, with distinct components handling context preparation, memory, summarization, and output interpretation.

LLM Orchestrator [Expand]

The central component for direct interaction with the Large Language Model, managing prompt submission, API calls, and raw response retrieval. It acts as the gateway for all LLM-driven operations and integrates basic memory.

Related Classes/Methods:

Context Management Module

Responsible for preparing and optimizing input context for the LLM, including token limit fitting, truncation, and integration of summarized or memorized information.

Related Classes/Methods:

Memory Management Module [Expand]

Manages the system's persistent and transient memory, storing past interactions, facts, or conversational history to provide crucial context for LLM prompts.

Related Classes/Methods:

Condition Evaluation Module

Interprets and evaluates specific conditions or logical statements based on the LLM's output, parsing structured data to determine true/false outcomes or extract values.

Related Classes/Methods:

GPT Selection Module

Facilitates decision-making and option selection processes by formatting choices for the LLM and parsing the LLM's chosen option(s) from its response.

Related Classes/Methods:

Summarization Module

Processes and condenses large blocks of text into shorter, manageable summaries, vital for handling extensive documents or conversations that exceed LLM token limits.

Related Classes/Methods:

LLM (External)

Represents the external Large Language Model service that the LLM Orchestrator interacts with for generating predictions and responses. This component is external to the thinkgpt codebase.

Related Classes/Methods: None

FAQ