module owlite.calib.mse_calibrator
class MSECalibrator
MSE Calibrator Class.
The MSECalibrator solves the $$ step_size $$ that minimizeds the mean squared error (MSE) between the original data and its quantized representation. When the original data $$ X $$ and the quantized representation $$ X_{quant} $$ is given, the optimal $$ step_size $$ is calculated as follow:
$$ step_size = \underset {step_size}{\operatorname{argmax}} \, {\left\| X - X{quant} \right\|}^22 $$
This approach minimizes the mean squared error between two data.
method __init__
python
__init__(quantizer: 'FakeQuantizer')
method update
python
update() → None
Update step_size using "mse
".
Updated: 2024-06-13T23:42:41