pytorch_quantization.calib¶
pytorch_quantization.calib provides Calibrator classes that
collect data statistics and determine pytorch_quantization parameters.
MaxCalibrator¶
- class pytorch_quantization.calib.MaxCalibrator(num_bits, axis, unsigned, track_amax=False)[source]¶
Max calibrator, tracks the maximum value globally
- Parameters
calib_desc – A MaxCalibDescriptor.
num_bits – An integer. Number of bits of quantization.
axis – A tuple. see QuantDescriptor.
unsigned – A boolean. using unsigned quantization.
- Readonly Properties:
amaxs: A list of amax. Numpy array is saved as it is likely to be used for some plot.
- collect(x)[source]¶
Tracks the absolute max of all tensors
- Parameters
x – A tensor
- Raises
RuntimeError – If amax shape changes
HistogramCalibrator¶
- class pytorch_quantization.calib.HistogramCalibrator(num_bits, axis, unsigned, num_bins=2048, grow_method=None, skip_zeros=False, torch_hist=False)[source]¶
Unified histogram calibrator
- Histogram will be only collected once. compute_amax() performs entropy, percentile, or mse
calibration based on arguments
- Parameters
num_bits – An integer. Number of bits of quantization.
axis – A tuple. see QuantDescriptor.
unsigned – A boolean. using unsigned quantization.
num_bins – An integer. Number of histograms bins. Default 2048.
grow_method – A string. DEPRECATED. default None.
skip_zeros – A boolean. If True, skips zeros when collecting data for histogram. Default False.
torch_hist – A boolean. If True, collect histogram by torch.histc instead of np.histogram. If input tensor is on GPU, histc will also be running on GPU. Default False.
- compute_amax(method: str, *, stride: int = 1, start_bin: int = 128, percentile: float = 99.99)[source]¶
Compute the amax from the collected histogram
- Parameters
method – A string. One of [‘entropy’, ‘mse’, ‘percentile’]
- Keyword Arguments
stride – An integer. Default 1
start_bin – An integer. Default 128
percentils – A float number between [0, 100]. Default 99.99.
- Returns
amax – a tensor