LLM MEM

LLM-MEM ?

An Enhanced Language Model with Semantic Memory Storage

A Python-based implementation that augments Language Models with persistent semantic memory using vector embeddings and similarity search.

? Features

• Semantic chunking of conversations using NLTK
• Vector embeddings generation using Sentence Transformers
• Cosine similarity-based context retrieval
• Persistent memory storage using vector database
• Automatic context injection for more coherent responses
• Smart text chunking with code block preservation

Technical Stack

• sentence-transformers: For generating semantic embeddings
• nltk: Natural Language Processing toolkit for text manipulation
• Custom LLM integration
• Vector-store for storing and retrieving embeddings
• REST API endpoints for vector operations

? Prerequisites

• sentence-transformers
• nltk
• requests
• google-generativeai

Quick Start

1. Clone the repository:

git clone https://github.com/Abhigyan126/LLM-MEM.git
cd LLM-MEM

2. Install dependencies:

pip install sentence-transformers nltk requests google-generativeai

3. create .env file

   key = your_llm_key

4. Run the main script:

python main.py

How It Works

1. Text Processing: Incoming messages are processed and cleaned
2. Embedding Generation: Converts text into vector embeddings
3. Semantic Search: Finds relevant previous contexts using cosine similarity
4. Context Integration: Merges relevant history with current query
5. Response Generation: Generates response using LLM with enhanced context
6. Memory Storage: Stores new interactions for future reference

? Configuration

The system uses several configurable parameters:

• Maximum chunk length: 2000 characters
• Minimum chunk size: 100 characters
• Similarity threshold: 0.5
• Default nearest neighbors: 10