Expand description
§Incin Macros
incin-macros provides procedural macros that form the ergonomic frontend for the Incin framework.
Because Incin relies heavily on compile-time type verification using typenum, manipulating raw types can be extremely verbose (e.g., Prod<UInt<UTerm, B1>, ...>).
These macros hide that complexity, allowing developers to write clean, intuitive syntax while the macros generate the underlying type boundaries.
§Provided Macros
s![...]: Defines a shape directly from integer literals (e.g.,s![1, 3, 224, 224]).i![...]: Defines runtime indexing/slicing arguments for tensors (e.g.,i![1..3, .., -1]).idx![...]: Defines an advanced type-level slicing target.#[module]: Derives neural network module traits automatically for structs, similar to#[derive(Module)]but specifically tailored for Incin.model!("model.onnx", Name): Compiles an ONNX model file into a fully-typed Rust module structurally matching the ONNX graph.import_model!("model.onnx", Name): Compiles an ONNX model file into a fully-typed Rust module structurally matching the ONNX graph.
Macros§
- axis
- Build an arbitrary static or runtime axis selector.
- i
- Build runtime indexing and slicing expressions for
Tensor::get. - idx
- A macro to construct index and slicing arguments for tensor reshaping and subsetting.
- import_
model - Attribute-style entry:
#[distributed_main]selects and wraps a process-rank main. - model
- Expands a supported
.onnxgraph or imports.safetensorsmetadata. - s
- A macro to construct static Tensor shapes ergonomically in the type system.
- shape
- Builds a
Tensordirectly from a literal, inferring shape and (unless told otherwise) dtype from how it’s written - the macro analogue ofvec![...]. - tensor
- Procedural macro
tensor!building a typed tensor literal with static shape checking.
Attribute Macros§
- module
- The
#[module]attribute automatically derives composable module capabilities for custom neural network structs.