From 91973111425bd04fdd841a0a67ca1bb6fe41c0dd Mon Sep 17 00:00:00 2001 From: "Wang, Mengni" Date: Thu, 25 Jul 2024 18:06:36 +0800 Subject: [PATCH] Update quantization.md Signed-off-by: Wang, Mengni --- docs/quantization.md | 8 -------- 1 file changed, 8 deletions(-) diff --git a/docs/quantization.md b/docs/quantization.md index 3831e453f..28e34fa66 100644 --- a/docs/quantization.md +++ b/docs/quantization.md @@ -63,14 +63,6 @@ Sometimes the reduce_range feature, that's using 7 bit width (1 sign bit + 6 dat > Activation (uint8) + Weight (int8) is recommended for performance on x86-64 machines with AVX2 and AVX512 extensions. -#### Quantization Scheme -+ Symmetric Quantization - + int8: scale = 2 * max(abs(rmin), abs(rmax)) / (max(int8) - min(int8) - 1); zero_point = 0 - + uint8: scale = 2 * max(abs(rmin), abs(rmax)) / (max(uint8) - min(uint8)); zero_point = 0 -+ Asymmetric Quantization - + int8: scale = (rmax - rmin) / (max(int8) - min(int8)); zero_point = round(min(int8) - rmin / scale) - + uint8: scale = (rmax - rmin) / (max(uint8) - min(uint8)); zero_point = round(min(uint8) - rmin / scale) - #### Reference + MLAS: [MLAS Quantization](https://github.com/microsoft/onnxruntime/blob/master/onnxruntime/python/tools/quantization/onnx_quantizer.py)