genxcode

genxcode generates boilerplate code from templates.

Use it to create starter code for templates like PyTorch, CUDA, and Hugging Face, then customize the generated project for your use case.

Install

pip install genxcode

CLI Workflow

Run:

genxcode init pytorch

Or omit the template argument and it prompts in the terminal:

Enter the name of template:

Accepted inputs include canonical names and aliases:

• pytorch
• cuda
• huggingface
• hf
• llm

hf resolves to the Hugging Face template.

If you enter pytorch, genxcode writes genxcode.yaml like this:

# Generated by `genxcode init pytorch`
template: pytorch
name: pytorch
arch: mlp
input_dim: 784
output_dim: 10
hidden:
  - 256
  - 128
activation: relu
metrics:
  - accuracy
  - loss
prod: false
lr: 0.001
epochs: 10
batch_size: 32
device: cpu

Edit the values you want, then generate the project:

genxcode apply

This creates a folder named after name and writes the boilerplate inside it. With the default config above, the output lands in ./pytorch/.

Output:

pytorch/
    src/
        __init__.py
        config.py
        models.py
        data.py
        train.py
        predict.py
    main.py
    requirements.txt

Available templates:

• pytorch: PyTorch training scaffold
• cuda: CUDA kernel scaffold
• huggingface: Hugging Face LLM scaffold

template: is required in genxcode.yaml, so apply always knows which scaffold to render.

To upgrade the installed package in the current environment:

genxcode update

Then run the generated boilerplate:

python main.py