aimet_tensorflow.keras.bn_reestimation¶
Top-level API
- aimet_tensorflow.keras.bn_reestimation.reestimate_bn_stats(model, bn_re_estimation_dataset, bn_num_batches=100)[source]¶
top level api for end user directly call
- Parameters:
model (
Model) – tf.keras.Modelbn_re_estimation_dataset (
DatasetV2) – Training datasetbn_num_batches (
int) – The number of batches to be used for reestimation
- Return type:
Handle- Returns:
Handle that undos the effect of BN reestimation upon handle.remove()
- aimet_tensorflow.keras.batch_norm_fold.fold_all_batch_norms_to_scale(sim)[source]¶
Fold all batch_norm layers in a model into the quantization scale parameter of the corresponding conv layers
- Parameters:
sim (
QuantizationSimModel) – QuantizationSimModel to be folded- Return type:
List[Tuple[QcQuantizeWrapper,QcQuantizeWrapper]]- Returns:
A list of pairs of layers [(Conv/Linear, BN layer that got folded)]