Skip to content

mini-kit

Minimal, hackable building blocks for deep learning projects. Each layer is small enough to read, tweak, or drop into your own codebase.

Quick start

pip install mini-kit

You now have three helpers that work together:

  • mini.config loads plain-Python configs with parents, parameters, and CLI overrides.
  • mini.builder turns dictionaries into Python objects via registries or dotted import paths.
  • mini.trainer runs a lightweight PyTorch training loop with hooks for logging and checkpointing.

Minimal example

# 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},
}

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 Python config files; apply_overrides tweaks nested keys with CLI-friendly syntax.
  • build resolves "type" from a registry or import path and wires dependencies for you.

Packages

  • mini.config: Python-first configs with parents, parameters, overrides, and snapshots.
  • mini.builder: Registry-based factory that instantiates nested dictionaries.
  • mini.trainer: Minimal PyTorch trainer with hooks for progress, checkpointing, and logging.