Crate tract_core

Tract

Tiny, no-nonsense, self contained, portable TensorFlow and ONNX inference.

Example

use tract_core::internal::*;

// build a simple model that just add 3 to each input component
let mut model = InferenceModel::default();

let input = model.add_source_default("input").unwrap();
let three = model.add_const("three".to_string(), tensor0(3f32)).unwrap();
let add = model.add_node_default("add".to_string(),
    tract_core::ops::math::add::bin()
    ).unwrap();

model.add_edge(OutletId::new(input, 0), InletId::new(add, 0)).unwrap();
model.add_edge(OutletId::new(three, 0), InletId::new(add, 1)).unwrap();
model.auto_outputs().unwrap();

// We build an execution plan. Default inputs and outputs are inferred from
// the model graph.
let plan = SimplePlan::new(&model).unwrap();

// run the computation.
let input = ndarray::arr1(&[1.0f32, 2.5, 5.0]);
let mut outputs = plan.run(tvec![input.into()]).unwrap();

// take the first and only output tensor
let mut tensor = outputs.pop().unwrap();

// unwrap it as array of f32
let tensor = tensor.to_array_view::<f32>().unwrap();
assert_eq!(tensor, ndarray::arr1(&[4.0, 5.5, 8.0]).into_dyn());

While creating a model from Rust code is useful for testing the library, real-life use-cases will usually load a TensorFlow or ONNX model using tract-tensorflow or tract-onnx crates.

Re-exports

pub extern crate error_chain;

pub extern crate ndarray;

pub use crate::errors::*;

Modules

analyser	Model graph type inference.
broadcast	N-way tensor broadcast
datum	Tensor is the main data container for tract
dim	Extended dimension support
errors	error_chain generated types
framework	Enforce consistent API between the implemented Frameworks importers.
internal	This prelude is meant for code extending tract (like implementing new ops).
macros
model	B
ops	Ops
plan
prelude	This prelude is meant for code using tract.
pulse
tensor	Tensor, tract main data object of interest.

Macros

args_1
args_2
args_3
args_4
args_5
assert_backward	Asserts that backward inference results work as expected.
assert_close
assert_forward	Asserts that forward inference results work as expected.
bin_to_super_type
boxed_new
canonic
dimfact	Constructs a dimension fact.
dispatch_copy
dispatch_datum
dispatch_floatlike
dispatch_model
dispatch_model_no_pulse
dispatch_numbers
dispatch_signed
element_map
element_map_inplace
element_map_move
element_map_with_params
impl_op_same_as
inference_op_as_op
not_a_typed_op
op_as_typed_op
shapefact	Constructs a shape fact.
to_typed
tvec
typed_op_as_op
typefact	Constructs a type fact.
unwrap_or_none	Tries to unwrap an option, or returns Ok(None) otherwise.
valuefact	Constructs an value fact.
wrap