Struct PrecisionConverter 

pub struct PrecisionConverter { /* private fields */ }

Precision converter for array operations

Implementations

impl PrecisionConverter

pub fn new(config: PrecisionConfig) -> Self

Create a new precision converter

pub fn convert_and_compute_1d( &self, data: &Array1<f32>, op: impl Fn(&Array1<f32>) -> Array1<f32>, ) -> InferenceResult<Array1<f32>>

Convert FP32 array to reduced precision and back (for inference)

pub fn to_fp16_1d(&self, data: &Array1<f32>) -> Vec<f16>

Convert FP32 array to FP16

pub fn from_fp16_1d(&self, data: &[f16]) -> Array1<f32>

Convert FP16 array to FP32

pub fn to_bf16_1d(&self, data: &Array1<f32>) -> Vec<bf16>

Convert FP32 array to BF16

pub fn from_bf16_1d(&self, data: &[bf16]) -> Array1<f32>

Convert BF16 array to FP32

pub fn to_fp16_2d(&self, data: &Array2<f32>) -> Vec<f16>

Convert 2D FP32 array to FP16

pub fn from_fp16_2d( &self, data: &[f16], shape: (usize, usize), ) -> InferenceResult<Array2<f32>>

Convert FP16 to 2D FP32 array

pub fn to_bf16_2d(&self, data: &Array2<f32>) -> Vec<bf16>

Convert 2D FP32 array to BF16

pub fn from_bf16_2d( &self, data: &[bf16], shape: (usize, usize), ) -> InferenceResult<Array2<f32>>

Convert BF16 to 2D FP32 array

pub fn config(&self) -> &PrecisionConfig

Get the configuration