How adlarx9/neural-network works
This is a from-scratch neural-network learning library with example programs for image generation, text prediction, number recognition, and shape learning. A typical workflow is to prepare training data, build a model from layers or a predefined architecture, train it, and use a script or visualization to inspect results. The graph preserves the documented model families and examples; unsampled implementations are represented at their exact tree paths, but their internal wiring is not inferred.
Main entry points: rcmha.py Causal Attention · trainer.py Trainer · scripts Example Programs · basics Layers And Blocks
Components 21
Core Modeling
- Layers And Blocks
core/basics - Activations
core/activation - Parameters And Norms
core/parameterized - Flow Operations
core/flowmakers - Tensor Transforms
core/transform - Loss Functions
core/loss
Model Architectures
- MLP And CNN
core/network - LSTM Network
core/network/lstm.py - LLaMA Network
core/network/llama.py - Diffusion Models
core/network/diffusion - Causal Attention
core/block/rcmha.py
Text Processing
- Text Network
core/text/text_network.py - Tokenizers
core/text - Text Embedding
core/parameterized/embedding.py
Training And Persistence
- Training Data
core/training/data.py - Trainer
core/training/trainer.py - Model Storage
data
Examples And Visualization
- Program Entry
main.py - Example Programs
scripts - Result Visualizations
graphics
Other components
- Program Author
Connections 16
- Program Author → Program Entry: starts
README.md - Program Entry → Example Programs: dispatches examples
main.py - Example Programs → Diffusion Models: trains image model
README.md - Example Programs → Result Visualizations: shows results
README.md - Trainer → Training Data: uses data
core/training/trainer.py - Trainer → Loss Functions: uses loss
core/training/trainer.py - Trainer → Layers And Blocks: trains layers
core/training/trainer.py - Text Network → Layers And Blocks: extends block
core/text/text_network.py - Text Network → Text Embedding: embeds input
core/text/text_network.py - Text Embedding → Tokenizers: tokenizes text
core/parameterized/embedding.py - Causal Attention → Flow Operations: uses operations
core/block/rcmha.py - Causal Attention → Tensor Transforms: uses transforms
core/block/rcmha.py - Causal Attention → Activations: uses softmax
core/block/rcmha.py - Causal Attention → Parameters And Norms: uses projections
core/block/rcmha.py - Layers And Blocks → Activations: hosts layers
core/activation/activation.py - Model Storage → Program Entry: loads models