Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

14 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

SynapseGrid

SynapseGrid is an autonomous neural architect system designed to streamline the process of building, training, and explaining deep learning models. it acts as an AI-powered assistant that analyzes your task, designs the optimal neural architecture, and generates production-ready pytorch code

features

  • intelligent analysis: automatically analyzes task descriptions (e.g., "classify images", "predict housing prices") and data types to determine the best modeling approach

  • automated architecture design: designs neural networks (CNNs, MLPs, transformers) tailored to your specific task and compute budget

  • code generation: generates a complete, runnable pytorch project structure including:

    • model.py: the model definition

    • train.py: training loop with validation and logging

    • hpo.py: hyperparameter optimization scripts (using optuna)

    • explain.py: explainable AI (XAI) scripts using SHAP

  • explainable AI (XAI): built-in support for generating explanation scripts to understand model decisions (SHAP/GradCAM)

  • deployment ready: tools to package and deploy trained models

installation

  1. clone the repository:

    git clone https://github.com/Youcef3939/SynapseGrid.git 
    cd SynapseGrid
  2. install dependencies:

    pip install -r requirements.txt

usage

SynapseGrid provides a powerful command line interface (CLI)

1. build a model

to generate a project for a specific task:

python -m synapse_grid.cli build --task "classify images of cats and dogs" --data ./data/cats_dogs --compute medium

options:

  • --task: description of what you want to solve

  • --data: path to your dataset

  • --compute: compute budget (low, medium, high)

  • --hpo: enable hyperparameter optimization (flag)

2. deploy a model

to prepare a trained model for deployment:

python -m synapse_grid.cli deploy --model-path ./output/checkpoints/model.pth

generated project structure

when you run build, SynapseGrid creates a directory with:

output/
├── model.py       # the neural network architecture
├── train.py       # training script
├── explain.py     # XAI explanation script
├── hpo.py         # HPO script
├── data/          # data directory
└── checkpoints/   # model checkpoints

dataflow

alt text

interface architecture

alt text alt text

pipeline architecture

alt text

running tests

to run the internal test suite:

pytest tests/

this project is designed for researchers, engineers, and teams who need fast, reliable neural architectures without spending hours tuning models by hand<3

About

an autonomous AI architect that designs and trains neural networks from a simple task description

Resources

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages