Quickstart

Curated first-tour examples for TorchLean.

This source umbrella teaches the primitives a new user needs before opening the larger model examples:

• typed tensors and runtime arithmetic,
• one tensor-function and one model-state gradient,
• an end-to-end MLP training command,
• typed model execution at a caller-selected binary precision,
• an explicit typed SGD loop that preserves the selected precision, and
• small proofs over TorchLean definitions.

Data loading, CNNs, larger models, editor widgets, advanced autograd transforms, interoperability, and verification live in their focused example directories. Quickstart is a learning path, not a model catalog. The torchlean executable imports the three runnable quickstarts directly; precision and proof modules are only included in this source umbrella.