Skip to main content

QuantizedTensorInt4

Type Alias QuantizedTensorInt4 

Source
pub type QuantizedTensorInt4 = QuantizedTensorGeneric<Int4Range>;
Expand description

An INT4 quantized tensor with optional per-channel parameters and bit packing.

Values are stored in the range [-8, 7]. Call pack to compress two values into one byte for 2× storage savings.

Aliased Type§

pub struct QuantizedTensorInt4 { /* private fields */ }