mini-kit

Minimal, hackable building blocks for deep learning projects. The goal is to keep every layer small enough that you can read it, tweak it, or drop it into your own codebase.

Quick Start

pip install mini-kit

You now have three tiny helpers that play nicely together:

1. mini.config loads plain-Python configs. Start with a single dictionary, then grow into composable templates and parent chains without learning a new DSL.
2. mini.builder turns dictionaries into Python objects. Supports both registry shortcuts and fully qualified import paths.
3. mini.trainer provides a lightweight training framework with hooks for logging, checkpointing, and customization.

The example below shows mini.config and mini.builder in action.

# configs/model.py
config = {
    "model": {
        "type": "Classifier",
        "encoder": {"type": "Encoder", "channels": 64},
        "head": {"type": "torch.nn.Linear", "*": [64, 10]},
    },
    "optimizer": {"type": "torch.optim.AdamW", "lr": 3e-4},
}

# main.py
from mini.builder import register, build
from mini.config import apply_overrides, load

@register()
class Encoder:
    def __init__(self, channels: int):
        self.channels = channels

@register()
class Classifier:
    def __init__(self, encoder, head):
        self.encoder = encoder
        self.head = head

cfg = load("configs/model.py")
cfg = apply_overrides(cfg, ["optimizer.lr=1e-3", "model.encoder.channels=128"])

model = build(cfg["model"])
optimizer = build(cfg["optimizer"])

• load executes configs/model.py; keep a simple config = {...} for small projects, or swap to def config(...): and parents = [...] when you need templates and composition.
• apply_overrides allows for painless command-line overrides: tweak nested keys with a short-hand syntax: optimizer.lr=..., append with +=, or drop entries with !=.
• build looks at the "type" key, grabs the right constructor (from the registry or import path), and wires up dependencies for you.

More details are in each subpackage's own README:

• mini.config: README
• mini.builder: README
• mini.trainer: README