dnrops
/
burn
mirror of https://github.com/tracel-ai/burn.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Fix symmetric quantization

This commit is contained in:
Guillaume Lagrange 2024-06-06 10:27:31 -04:00
parent 96b46b59d0
commit af94ad0a5e
1 changed files with 6 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
crates/burn-tensor/src/tensor/quantization.rs
View File

@ -116,9 +116,10 @@ impl<E: Float, Q: PrimInt> Quantization<E, Q> for SymmetricQuantization<E, Q> {
let b = E::from(Q::max_value()).unwrap();
let a = b.neg();
// Compute scale to convert a floating point value in range `[alpha, beta]` to the quantized range
// Compute scale to convert a floating point value in range `[-alpha, alpha]` to the quantized range
let alpha = alpha.abs().max(beta.abs());
Self {
scale: (beta - alpha) / (b - a),
scale: (alpha + alpha) / (b - a),
_q: PhantomData,
}
}
@ -173,6 +174,7 @@ impl QuantizationStrategy {
#[cfg(test)]
mod tests {
use super::*;
use alloc::vec;
#[test]
fn test_int8_affine_quantization() {
@ -193,8 +195,8 @@ mod tests {
#[test]
fn test_int8_symmetric_quantization() {
let x: [f32; 4] = [-1.8, -1.0, 0.0, 0.5];
let expected_q = vec![-127, -110, 0, 55];
let expected_d = vec![-1.15, -0.996063, 0.0, 0.4980315];
let expected_q = vec![-127, -71, 0, 35];
let expected_d = vec![-1.8, -1.0062993, 0.0, 0.496063];
let symmetric = QuantizationStrategy::Int8Symmetric(SymmetricQuantization::new(-1.8, 0.5));