LEARN

Examples Curriculum

The Laco examples curriculum is a set of progressively more complex configuration examples demonstrating the full API. All examples live under sources/laco/examples/ and are exercised by tests/test_examples.py.

Tier 0–1: Foundations

Module	What it demonstrates
`mlp.py`	`L.call`, `L.params`, `L.repeat`, `L.OrderedDict`
`linear_regression.py`	`L.call`, `L.partial`, `L.params`
`cnn_classifier.py`	Stacked stages with `L.repeat`
`text_classifier.py`	`L.required[T]()` for mandatory fields

Minimal example: linear regression

import laco.language as L
from torch import nn, optim

@L.params
class hps:
    in_features: int = 8
    out_features: int = 1
    learning_rate: float = 1e-2

model = L.call(nn.Linear, root=True)(
    in_features=hps.in_features,
    out_features=hps.out_features,
)
optimizer = L.partial(optim.SGD)(lr=hps.learning_rate)

Load and instantiate:

cfg = laco.load("configs://examples/linear_regression.py#model")
model = laco.instantiate(cfg)

Override from CLI:

laco compose configs://examples/linear_regression.py hps.in_features=16

Tier 0–1: Typed-group variants

Each Tier 0–1 example has a sibling in examples/typed/ that uses the typed-group API (L.Group, @L.config, L.slot, L.bind).

Module	Changes
`typed/mlp.py`	`ActivationGroup(L.Group[nn.Module])` for swappable activations
`typed/linear_regression.py`	`OptimGroup(L.Group[Optimizer])`
`typed/text_classifier.py`	`@L.config` schema, `L.required[int]()`

Typed-group example: MLP

class ActivationGroup(L.Group[nn.Module]):
    relu = L.call(nn.ReLU)()
    gelu = L.call(nn.GELU)()

@L.config
class MLPSchema:
    dim_in: int = 128
    activation: nn.Module = L.slot(ActivationGroup)

defaults = L.Defaults(
    L.self_,
    L.bind(MLPSchema.activation, ActivationGroup.relu),  # default to ReLU
)

Override from CLI: laco compose examples/typed/mlp.py activation=gelu

Tier 2: Building Blocks

Module	What it demonstrates
`blocks/residual.py`	Residual connection, `L.call` factories
`blocks/transformer.py`	Multi-head attention + FFN layers
`blocks/decoder.py`	Autoregressive decoder

Tier 3–4: Vision & Language Models

Large model configs under examples/models/.

Tier 5: Ecosystem Integrations

Module	Integration
`integrations/lightning_module.py`	PyTorch Lightning
`integrations/transformers_qa.py`	HuggingFace Transformers
`integrations/tensordict_module.py`	TensorDict

Tier 6: End-to-end Pipelines

Module	What it demonstrates
`pipelines/mnist_train.py`	MNIST training loop with `@L.task`
`pipelines/clm_finetune.py`	Causal-LM fine-tuning pipeline

Running the MNIST pipeline

# Smoke run (1 step, no real data needed):
python -m laco.examples.pipelines.mnist_train

# 10 steps on real MNIST (downloads ~11 MB):
python -m laco.examples.pipelines.mnist_train hps.num_steps=10

Migrate from argparse

Examples Curriculum