Guides#

These guides cover the working surfaces of LEAPP in more depth. Read them in order or jump to the topic you need.

Node patterns

Distributed inputs, method shorthand, and nested data structures.

Node patterns
Export configuration

Backend selection, TorchScript vs ONNX, and bringing pre-compiled models into the graph.

Export configuration
State capture and feedback

Explicit state tensors, module buffer tracking, and automatic feedback detection for recurrent or re-entered graph state.

State capture and feedback
Buffers and constant tensors

Registered module buffers, static outputs, embedded constants, and manual tensor-copy tag preservation.

Buffers and constant tensors
Debugging

Dry-run modes, selective non-traced nodes, verbose logs, and full FX graph inspection.

Debugging
Runtime and validation

Per-node validation, cached re-entry examples, and the Python InferenceManager for end-to-end smoke tests.

Runtime and validation
Semantic data annotation

Use TensorSemantics to describe what tensors represent and name individual tensor elements.

Semantic data annotation