This project aims to create an external memory module for language models, enabling agent-like capabilities for research and learning. It targets users seeking to enhance LLM memory, logic, and interpretability, offering a structured approach to knowledge organization and retrieval.

How It Works

The system structures information by extracting concepts from text, creating human-interpretable, natural language embeddings for each piece of data. These embeddings are then clustered hierarchically, forming a knowledge graph. When a question is posed, the system generates a tailored embedding for it, identifies relevant information within the graph, and uses this context to prompt the language model for an answer.

Quick Start & Requirements

• Install: pip install -r requirements.txt
• Prerequisites: Python 3.x, OpenAI API key.
• Demo: https://github.com/tomhartke/knowledge-graph-from-GPT/blob/main/demo.ipynb

Highlighted Details

• Creates human-interpretable knowledge graphs where embedding dimensions are natural language concepts.
• Embeddings are tailored to the local structure of learned knowledge, not global datasets.
• Supports hypothesis generation by clustering existing questions to inspire new ones.
• Includes an interactive UI for exploring topics, generating, and answering questions.

Maintenance & Community

The project is maintained by tomhartke. There are no explicit community channels or roadmap links provided in the README.

Licensing & Compatibility

The repository does not explicitly state a license. Users should exercise caution regarding commercial use or integration with closed-source projects.

Limitations & Caveats

The clustering algorithm is described as "hacked together" and not polished, though functional. The project is a proof-of-concept, with significant future work required to achieve agent-like capabilities. Handling of non-text data, synonyms, and very short/long texts are noted as foreseen issues.