graph LR
LLM_Predictor["LLM Predictor"]
LLM_Generator["LLM Generator"]
LLM_Predictor -- "offers distinct functionality from" --> LLM_Generator
The thinkgpt.llm subsystem provides core functionalities for interacting with Large Language Models, primarily through two distinct interfaces: LLM Predictor and LLM Generator. The LLM Predictor serves as a streamlined entry point for direct, stateless LLM queries, focusing on preparing the input and invoking the underlying LLM execution mechanism. In contrast, the LLM Generator offers a more sophisticated interface, incorporating context management capabilities to facilitate conversational or stateful interactions with the LLM. Both components independently leverage a shared internal execution chain to process requests and retrieve LLM responses, ensuring flexibility in how external systems can interact with the LLM while maintaining a consistent backend for processing.
Serves as a direct, simplified interface for initiating LLM predictions. Its primary role is to prepare the input for the underlying LLM execution chain and trigger the prediction process, acting as a straightforward wrapper for basic prediction requests.
Related Classes/Methods:
Manages the core logic for generating LLM responses, including sophisticated context management. It prepares prompts, potentially incorporating conversational history or other contextual information via internal mechanisms like self.remember, and then dispatches requests to the underlying LLM execution chain. It handles the immediate retrieval of the raw LLM output and any initial processing related to context.
Related Classes/Methods: