Self-AI-Trainer: Real-Time Neural Network Visualization
Training a local LLM on your own data with a live visualization of the neural network changing in real time.
Motivation
The idea behind this project is simple: you should be able to see what your model is learning while it learns. Most training pipelines give you loss curves and validation metrics, but they don't show what's happening inside the network. This project aims to change that.
What It Does
- Data ingestion: Feed your own text data (notes, writing samples, documents) into the training pipeline
- Real-time visualization: Watch the neural network weights and gradients update in real time during training
- Interactive chat: After training, interact with the model through a chat interface tuned to your data
- Text humanization: Use the trained model to rewrite text in your own voice and style
Technical Approach
The system uses a local LLM (small enough to train on consumer hardware) with a custom training loop that emits weight/gradient snapshots at each step. These snapshots are streamed to a web-based visualization frontend that renders the network state in real time.
Current Status
This project is currently in progress. The training pipeline and visualization frontend are functional, but the chat interface and text humanization features are still being developed.
Source: GitHub Repository