truston::client::io

Struct InferInput

pub struct InferInput {
    pub input_name: String,
    pub input_shape: Vec<usize>,
    pub input_data: DataType,
}

Expand description

Represents a single input tensor for inference requests.

InferInput bundles together:

the input name (as expected by the model),
the input shape (tensor dimensions),
and the input data (wrapped in DataType).

§Examples

Creating manually:

use truston::client::io::{InferInput, DataType};

let input = InferInput::new(
    "input_tensor".into(),
    vec![2, 2],
    DataType::F32(vec![0.1, 0.2, 0.3, 0.4]),
);
assert_eq!(input.input_shape, vec![2, 2]);

Creating directly from an ndarray:

use ndarray::array;
use truston::client::io::InferInput;

let arr = array![[1.0f32, 2.0], [3.0, 4.0]].into_dyn();
let input = InferInput::from_ndarray("matrix_input", arr);
assert_eq!(input.input_shape, vec![2, 2]);

§Notes

MVP1: only ArrayD<T> with Vec<T>: IntoInferData is supported.
MVP2: future versions may support zero-copy or borrowed buffers for better performance.

Fields§

§input_name: String§input_shape: Vec<usize>§input_data: DataType

Implementations§

Source §

impl InferInput

Source

pub fn new( input_name: String, input_shape: Vec<usize>, input_data: DataType, ) -> Self

Examples found in repository ?

examples/random_tensor.rs (line 7)

5fn main() {
6    let dummy_input = DataType::F32(vec![5.0, 5.0, 5.0]);
7    let _input = InferInput::new("input_0".into(), vec![1, 3], dummy_input);
8}

Source

pub fn from_ndarray<T>(name: impl Into<String>, arr: ArrayD<T>) -> Self
where T: Clone + 'static, Vec<T>: IntoInferData,

Examples found in repository ?

examples/ndarray_coba.rs (line 9)

5fn main() {
6    // bikin tensor shape (2,10) → 2x10 matrix
7    let arr: ArrayD<f32> = ArrayD::zeros(IxDyn(&[2, 10]));
8
9    let infer_input = InferInput::from_ndarray("dynamic_input", arr);
10
11    println!("{:#?}", infer_input.input_data);
12}

More examples

Hide additional examples

examples/infer_test.rs (line 10)

8async fn run_infer(client: Arc<TritonRestClient>) -> Result<(), TrustonError> {
9    let arr_ids: ArrayD<i64> = ArrayD::zeros(IxDyn(&[32, 128]));
10    let input_ids = InferInput::from_ndarray("input_ids", arr_ids);
11
12    let arr_attention_mask: ArrayD<i64> = ArrayD::zeros(IxDyn(&[32, 128]));
13    let input_attention_mask = InferInput::from_ndarray("attention_mask", arr_attention_mask);
14
15    let inputs = vec![input_ids, input_attention_mask];
16
17    let _ = client.infer(inputs, "hierarchical_clf").await?;
18    Ok(())
19}