diff --git a/docs/.buildinfo b/docs/.buildinfo index a760069d2..eafc4f405 100644 --- a/docs/.buildinfo +++ b/docs/.buildinfo @@ -1,4 +1,4 @@ # Sphinx build info version 1 # This file hashes the configuration used when building these files. When it is not found, a full rebuild will be done. -config: 5cc6500fd7d8a78cea89aef3268de5c5 +config: 463b5d411b812fb296a8f7bff970d1cf tags: 645f666f9bcd5a90fca523b33c5a78b7 diff --git a/docs/_sources/api/api_docs/classes/BitWidthConfig.rst.txt b/docs/_sources/api/api_docs/classes/BitWidthConfig.rst.txt new file mode 100644 index 000000000..9b5cff54c --- /dev/null +++ b/docs/_sources/api/api_docs/classes/BitWidthConfig.rst.txt @@ -0,0 +1,14 @@ +:orphan: + +.. _ug-BitWidthConfig: + + +BitWidthConfig +========================== + +.. autoclass:: model_compression_toolkit.core.BitWidthConfig + +ManualBitWidthSelection +========================== + +.. autoclass:: model_compression_toolkit.core.common.quantization.bit_width_config.ManualBitWidthSelection diff --git a/docs/_sources/api/api_docs/classes/XQuantConfig.rst.txt b/docs/_sources/api/api_docs/classes/XQuantConfig.rst.txt new file mode 100644 index 000000000..79d34128a --- /dev/null +++ b/docs/_sources/api/api_docs/classes/XQuantConfig.rst.txt @@ -0,0 +1,14 @@ +:orphan: + +.. _ug-XQuantConfig: + +================================================ +XQuant Configuration +================================================ + +.. autoclass:: model_compression_toolkit.xquant.common.xquant_config.XQuantConfig + :members: + + + + diff --git a/docs/_sources/api/api_docs/index.rst.txt b/docs/_sources/api/api_docs/index.rst.txt index b349d3be4..0c4433163 100644 --- a/docs/_sources/api/api_docs/index.rst.txt +++ b/docs/_sources/api/api_docs/index.rst.txt @@ -49,6 +49,7 @@ core - :ref:`QuantizationConfig`: Module to configure the quantization process. - :ref:`QuantizationErrorMethod`: Select a method for quantization parameters' selection. - :ref:`MixedPrecisionQuantizationConfig`: Module to configure the quantization process when using mixed-precision PTQ. +- :ref:`BitWidthConfig`: Module to configure the bit-width manually. - :ref:`ResourceUtilization`: Module to configure resources to use when searching for a configuration for the optimized model. - :ref:`MpDistanceWeighting`: Mixed precision distance metric weighting methods. - :ref:`network_editor`: Module to modify the optimization process for troubleshooting. @@ -75,6 +76,13 @@ pruning - :ref:`PruningConfig`: Configuration for the pruning process (experimental). - :ref:`PruningInfo`: Information about the pruned model such as pruned channel indices, etc. (experimental). +xquant +=========== + +- :ref:`xquant_report_pytorch_experimental`: A function to generate an explainable quantization report for a quantized Pytorch model (experimental). +- :ref:`xquant_report_keras_experimental`: A function to generate an explainable quantization report for a quantized Keras model (experimental). + +- :ref:`XQuantConfig`: Configuration for the XQuant report (experimental). exporter ========= diff --git a/docs/_sources/api/api_docs/methods/xquant_report_keras_experimental.rst.txt b/docs/_sources/api/api_docs/methods/xquant_report_keras_experimental.rst.txt new file mode 100644 index 000000000..af6994d1d --- /dev/null +++ b/docs/_sources/api/api_docs/methods/xquant_report_keras_experimental.rst.txt @@ -0,0 +1,12 @@ +:orphan: + +.. _ug-xquant_report_keras_experimental: + + +================================================ +XQuant Report Keras +================================================ + +.. autofunction:: model_compression_toolkit.xquant.keras.facade_xquant_report.xquant_report_keras_experimental + + diff --git a/docs/_sources/api/api_docs/methods/xquant_report_pytorch_experimental.rst.txt b/docs/_sources/api/api_docs/methods/xquant_report_pytorch_experimental.rst.txt new file mode 100644 index 000000000..074db79d9 --- /dev/null +++ b/docs/_sources/api/api_docs/methods/xquant_report_pytorch_experimental.rst.txt @@ -0,0 +1,15 @@ +:orphan: + +.. _ug-xquant_report_pytorch_experimental: + + +================================================ +XQuant Report Pytorch +================================================ + +.. autofunction:: model_compression_toolkit.xquant.pytorch.facade_xquant_report.xquant_report_pytorch_experimental + + + + + diff --git a/docs/api/api_docs/classes/BitWidthConfig.html b/docs/api/api_docs/classes/BitWidthConfig.html new file mode 100644 index 000000000..ffb941926 --- /dev/null +++ b/docs/api/api_docs/classes/BitWidthConfig.html @@ -0,0 +1,140 @@ + + + + + + + + + + BitWidthConfig — MCT Documentation: ver 2.2.0 + + + + + + + + + + + + + + + + + +
+
+
+
+ +
+

BitWidthConfig

+
+
+class model_compression_toolkit.core.BitWidthConfig(manual_activation_bit_width_selection_list=None)
+

Class to manage manual bit-width configurations.

+
+
+manual_activation_bit_width_selection_list
+

A list of ManualBitWidthSelection objects defining manual bit-width configurations.

+
+
Type:
+

List[ManualBitWidthSelection]

+
+
+
+ +
+ +
+
+

ManualBitWidthSelection

+
+
+class model_compression_toolkit.core.common.quantization.bit_width_config.ManualBitWidthSelection(filter, bit_width)
+

Class to encapsulate the manual bit width selection configuration for a specific filter.

+
+
+filter
+

The filter used to select nodes for bit width manipulation.

+
+
Type:
+

BaseNodeMatcher

+
+
+
+ +
+
+bit_width
+

The bit width to be applied to the selected nodes.

+
+
Type:
+

int

+
+
+
+ +
+ +
+ + +
+
+
+
+ +
+
+ + + + \ No newline at end of file diff --git a/docs/api/api_docs/classes/DataGenerationConfig.html b/docs/api/api_docs/classes/DataGenerationConfig.html index 333a6514f..3f2a64060 100644 --- a/docs/api/api_docs/classes/DataGenerationConfig.html +++ b/docs/api/api_docs/classes/DataGenerationConfig.html @@ -7,7 +7,7 @@ - Data Generation Configuration — MCT Documentation: ver 2.1.0 + Data Generation Configuration — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Data Generation Configuration
    • @@ -45,7 +45,7 @@

    Navigation

    Data Generation Configuration

    -class model_compression_toolkit.data_generation.DataGenerationConfig(n_iter, optimizer, data_gen_batch_size, initial_lr, output_loss_multiplier, image_granularity=ImageGranularity.AllImages, scheduler_type=None, bn_alignment_loss_type=None, output_loss_type=None, data_init_type=None, layer_weighting_type=None, image_pipeline_type=None, image_normalization_type=None, extra_pixels=0, bn_layer_types=[], last_layer_types=[], clip_images=True, reflection=True)
    +class model_compression_toolkit.data_generation.DataGenerationConfig(n_iter, optimizer, data_gen_batch_size, initial_lr, output_loss_multiplier, image_granularity=ImageGranularity.AllImages, scheduler_type=None, bn_alignment_loss_type=None, output_loss_type=None, data_init_type=None, layer_weighting_type=None, image_pipeline_type=None, image_normalization_type=None, extra_pixels=0, bn_layer_types=[], last_layer_types=[], image_clipping=True)

    Configuration class for data generation.

    Initialize the DataGenerationConfig.

    @@ -64,11 +64,10 @@

    Navigation

  • layer_weighting_type (BNLayerWeightingType) – Type of layer weighting. Defaults to None.

  • image_pipeline_type (ImagePipelineType) – Type of image pipeline. Defaults to None.

  • image_normalization_type (ImageNormalizationType) – Type of image normalization. Defaults to None.

    • -

  • extra_pixels (int) – Extra pixels to add to the input image size. Defaults to 0.

    • +

  • extra_pixels (Union[int, Tuple[int, int]]) – Extra pixels to add to the input image size. Defaults to 0.

  • bn_layer_types (List) – List of BatchNorm layer types. Defaults to [].

  • last_layer_types (List) – List of layer types. Defaults to [].

    • -

  • clip_images (bool) – Flag to enable image clipping. Defaults to True.

    • -

  • reflection (bool) – Flag to enable reflection. Defaults to True.

    • +

  • image_clipping (bool) – Flag to enable image clipping. Defaults to True.

    • @@ -112,10 +111,11 @@

    OutputLossType
    class model_compression_toolkit.data_generation.OutputLossType(value)
    -

    An enum for choosing the output loss type:

    -

    NONE - No output loss is applied.

    -

    MIN_MAX_DIFF - Use min-max difference as the output loss.

    -

    REGULARIZED_MIN_MAX_DIFF - Use regularized min-max difference as the output loss.

    +

    An enum for choosing the output loss type: +NONE - No output loss is applied. +NEGATIVE_MIN_MAX_DIFF - Use the mean of the negative min-max difference as the output loss. +INVERSE_MIN_MAX_DIFF - Use mean of the 1/(min-max) difference as the output loss. +REGULARIZED_MIN_MAX_DIFF - Use regularized min-max difference as the output loss.

    @@ -148,8 +148,7 @@

    ImagePipelineType class model_compression_toolkit.data_generation.ImagePipelineType(value)

    An enum for choosing the image pipeline type for image manipulation:

    -

    RANDOM_CROP - Crop the images.

    -

    RANDOM_CROP_FLIP - Crop and flip the images.

    +

    SMOOTHING_AND_AUGMENTATION - Apply a smoothing filter, then crop and flip the images.

    IDENTITY - Do not apply any manipulation (identity transformation).

    @@ -212,7 +211,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Data Generation Configuration
    • diff --git a/docs/api/api_docs/classes/DefaultDict.html b/docs/api/api_docs/classes/DefaultDict.html index d1072f12b..a8805dda9 100644 --- a/docs/api/api_docs/classes/DefaultDict.html +++ b/docs/api/api_docs/classes/DefaultDict.html @@ -7,7 +7,7 @@ - DefaultDict Class — MCT Documentation: ver 2.1.0 + DefaultDict Class — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • DefaultDict Class
    • @@ -114,7 +114,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • DefaultDict Class
    • diff --git a/docs/api/api_docs/classes/FrameworkInfo.html b/docs/api/api_docs/classes/FrameworkInfo.html index c6d14bf20..517e94030 100644 --- a/docs/api/api_docs/classes/FrameworkInfo.html +++ b/docs/api/api_docs/classes/FrameworkInfo.html @@ -7,7 +7,7 @@ - FrameworkInfo Class — MCT Documentation: ver 2.1.0 + FrameworkInfo Class — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • FrameworkInfo Class
    • @@ -144,7 +144,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • FrameworkInfo Class
    • diff --git a/docs/api/api_docs/classes/GradientPTQConfig.html b/docs/api/api_docs/classes/GradientPTQConfig.html index 754871e39..75eae64d5 100644 --- a/docs/api/api_docs/classes/GradientPTQConfig.html +++ b/docs/api/api_docs/classes/GradientPTQConfig.html @@ -7,7 +7,7 @@ - GradientPTQConfig Class — MCT Documentation: ver 2.1.0 + GradientPTQConfig Class — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • GradientPTQConfig Class
    • @@ -78,7 +78,7 @@

    GPTQHessianScoresConfig Class
    -class model_compression_toolkit.gptq.GPTQHessianScoresConfig(hessians_num_samples=16, norm_scores=True, log_norm=True, scale_log_norm=False)
    +class model_compression_toolkit.gptq.GPTQHessianScoresConfig(hessians_num_samples=GPTQ_HESSIAN_NUM_SAMPLES, norm_scores=True, log_norm=True, scale_log_norm=False, hessian_batch_size=ACT_HESSIAN_DEFAULT_BATCH_SIZE)

    Configuration to use for computing the Hessian-based scores for GPTQ loss metric.

    Initialize a GPTQHessianWeightsConfig.

    @@ -88,6 +88,7 @@

    GPTQHessianScoresConfig Class index -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • GradientPTQConfig Class
    • diff --git a/docs/api/api_docs/classes/MixedPrecisionQuantizationConfig.html b/docs/api/api_docs/classes/MixedPrecisionQuantizationConfig.html index 4ecfcea5c..43bc8cce8 100644 --- a/docs/api/api_docs/classes/MixedPrecisionQuantizationConfig.html +++ b/docs/api/api_docs/classes/MixedPrecisionQuantizationConfig.html @@ -7,7 +7,7 @@ - MixedPrecisionQuantizationConfig — MCT Documentation: ver 2.1.0 + MixedPrecisionQuantizationConfig — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • MixedPrecisionQuantizationConfig
    • @@ -46,7 +46,7 @@

    Navigation

    Class to configure the quantization process of the model when quantizing in mixed-precision:

    -class model_compression_toolkit.core.MixedPrecisionQuantizationConfig(compute_distance_fn=None, distance_weighting_method=MpDistanceWeighting.AVG, num_of_images=32, configuration_overwrite=None, num_interest_points_factor=1.0, use_hessian_based_scores=False, norm_scores=True, refine_mp_solution=True, metric_normalization_threshold=1e10)
    +class model_compression_toolkit.core.MixedPrecisionQuantizationConfig(compute_distance_fn=None, distance_weighting_method=MpDistanceWeighting.AVG, num_of_images=MP_DEFAULT_NUM_SAMPLES, configuration_overwrite=None, num_interest_points_factor=1.0, use_hessian_based_scores=False, norm_scores=True, refine_mp_solution=True, metric_normalization_threshold=1e10, hessian_batch_size=ACT_HESSIAN_DEFAULT_BATCH_SIZE)

    Class with mixed precision parameters to quantize the input model.

    Parameters:
    @@ -60,6 +60,7 @@

    Navigation

  • norm_scores (bool) – Whether to normalize the returned scores for the weighted distance metric (to get values between 0 and 1).

  • refine_mp_solution (bool) – Whether to try to improve the final mixed-precision configuration using a greedy algorithm that searches layers to increase their bit-width, or not.

  • metric_normalization_threshold (float) – A threshold for checking the mixed precision distance metric values, In case of values larger than this threshold, the metric will be scaled to prevent numerical issues.

    • +

  • hessian_batch_size (int) – The Hessian computation batch size. used only if using mixed precision with Hessian-based objective.

    • @@ -94,7 +95,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • MixedPrecisionQuantizationConfig
    • diff --git a/docs/api/api_docs/classes/MpDistanceWeighting.html b/docs/api/api_docs/classes/MpDistanceWeighting.html index cfa96a715..ac03fdcef 100644 --- a/docs/api/api_docs/classes/MpDistanceWeighting.html +++ b/docs/api/api_docs/classes/MpDistanceWeighting.html @@ -7,7 +7,7 @@ - MpDistanceWeighting — MCT Documentation: ver 2.1.0 + MpDistanceWeighting — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • MpDistanceWeighting
    • @@ -83,7 +83,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • MpDistanceWeighting
    • diff --git a/docs/api/api_docs/classes/PruningConfig.html b/docs/api/api_docs/classes/PruningConfig.html index 1bec51965..47ff68d96 100644 --- a/docs/api/api_docs/classes/PruningConfig.html +++ b/docs/api/api_docs/classes/PruningConfig.html @@ -7,7 +7,7 @@ - Pruning Configuration — MCT Documentation: ver 2.1.0 + Pruning Configuration — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Pruning Configuration
    • @@ -159,7 +159,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Pruning Configuration
    • diff --git a/docs/api/api_docs/classes/PruningInfo.html b/docs/api/api_docs/classes/PruningInfo.html index 299a60f33..0516ddc4c 100644 --- a/docs/api/api_docs/classes/PruningInfo.html +++ b/docs/api/api_docs/classes/PruningInfo.html @@ -7,7 +7,7 @@ - Pruning Information — MCT Documentation: ver 2.1.0 + Pruning Information — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Pruning Information
    • @@ -122,7 +122,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Pruning Information
    • diff --git a/docs/api/api_docs/classes/QuantizationConfig.html b/docs/api/api_docs/classes/QuantizationConfig.html index ecff6b4e6..227aa69d1 100644 --- a/docs/api/api_docs/classes/QuantizationConfig.html +++ b/docs/api/api_docs/classes/QuantizationConfig.html @@ -7,7 +7,7 @@ - QuantizationConfig — MCT Documentation: ver 2.1.0 + QuantizationConfig — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • QuantizationConfig
    • @@ -46,7 +46,7 @@

    Navigation

    Class to configure the quantization process of the model:

    -class model_compression_toolkit.core.QuantizationConfig(activation_error_method=QuantizationErrorMethod.MSE, weights_error_method=QuantizationErrorMethod.MSE, relu_bound_to_power_of_2=False, weights_bias_correction=True, weights_second_moment_correction=False, input_scaling=False, softmax_shift=False, shift_negative_activation_correction=False, activation_channel_equalization=False, z_threshold=math.inf, min_threshold=MIN_THRESHOLD, l_p_value=2, linear_collapsing=True, residual_collapsing=True, shift_negative_ratio=0.05, shift_negative_threshold_recalculation=False, shift_negative_params_search=False, concat_threshold_update=False)
    +class model_compression_toolkit.core.QuantizationConfig(activation_error_method=QuantizationErrorMethod.MSE, weights_error_method=QuantizationErrorMethod.MSE, relu_bound_to_power_of_2=False, weights_bias_correction=True, weights_second_moment_correction=False, input_scaling=False, softmax_shift=False, shift_negative_activation_correction=True, activation_channel_equalization=False, z_threshold=math.inf, min_threshold=MIN_THRESHOLD, l_p_value=2, linear_collapsing=True, residual_collapsing=True, shift_negative_ratio=0.05, shift_negative_threshold_recalculation=False, shift_negative_params_search=False, concat_threshold_update=False)

    Class to wrap all different parameters the library quantize the input model according to.

    Parameters:
    @@ -113,7 +113,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • QuantizationConfig
    • diff --git a/docs/api/api_docs/classes/QuantizationErrorMethod.html b/docs/api/api_docs/classes/QuantizationErrorMethod.html index b77b373dc..e2c40702a 100644 --- a/docs/api/api_docs/classes/QuantizationErrorMethod.html +++ b/docs/api/api_docs/classes/QuantizationErrorMethod.html @@ -7,7 +7,7 @@ - QuantizationErrorMethod — MCT Documentation: ver 2.1.0 + QuantizationErrorMethod — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • QuantizationErrorMethod
    • @@ -85,7 +85,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • QuantizationErrorMethod
    • diff --git a/docs/api/api_docs/classes/ResourceUtilization.html b/docs/api/api_docs/classes/ResourceUtilization.html index 00b0f8a40..54f03591f 100644 --- a/docs/api/api_docs/classes/ResourceUtilization.html +++ b/docs/api/api_docs/classes/ResourceUtilization.html @@ -7,7 +7,7 @@ - ResourceUtilization — MCT Documentation: ver 2.1.0 + ResourceUtilization — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • ResourceUtilization
    • @@ -89,7 +89,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • ResourceUtilization
    • diff --git a/docs/api/api_docs/classes/XQuantConfig.html b/docs/api/api_docs/classes/XQuantConfig.html new file mode 100644 index 000000000..2375020ec --- /dev/null +++ b/docs/api/api_docs/classes/XQuantConfig.html @@ -0,0 +1,101 @@ + + + + + + + + + + XQuant Configuration — MCT Documentation: ver 2.2.0 + + + + + + + + + + + + + + + + + +
    +
    +
    +
    + +
    +

    XQuant Configuration

    +
    +
    +class model_compression_toolkit.xquant.common.xquant_config.XQuantConfig(report_dir, custom_similarity_metrics=None)
    +

    Configuration for generating the report. +It allows to set the log dir that the report will be saved in and to add similarity metrics +to measure between tensors of the two models.

    +

    Initializes the configuration for explainable quantization.

    +
    +
    Parameters:
    +
      +

    • report_dir (str) – Directory where the reports will be saved.

      • +

    • custom_similarity_metrics (Dict[str, Callable]) – Custom similarity metrics to be computed between tensors of the two models. The dictionary keys are similarity metric names and the values are callables that implement the similarity metric computation.

      • +
    +
    +
    +
    + +
    + + +
    +
    +
    +
    + +
    +
    + + + + \ No newline at end of file diff --git a/docs/api/api_docs/index.html b/docs/api/api_docs/index.html index cf66594ae..292fc010a 100644 --- a/docs/api/api_docs/index.html +++ b/docs/api/api_docs/index.html @@ -7,7 +7,7 @@ - API Docs — MCT Documentation: ver 2.1.0 + API Docs — MCT Documentation: ver 2.2.0 @@ -35,7 +35,7 @@

    Navigation

  • previous |
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • API Docs
    • @@ -88,6 +88,7 @@

    core¶<

  • QuantizationConfig: Module to configure the quantization process.

  • QuantizationErrorMethod: Select a method for quantization parameters’ selection.

  • MixedPrecisionQuantizationConfig: Module to configure the quantization process when using mixed-precision PTQ.

    • +

  • BitWidthConfig: Module to configure the bit-width manually.

  • ResourceUtilization: Module to configure resources to use when searching for a configuration for the optimized model.

  • MpDistanceWeighting: Mixed precision distance metric weighting methods.

  • network_editor: Module to modify the optimization process for troubleshooting.

    • @@ -114,6 +115,14 @@

    pruningPruningInfo: Information about the pruned model such as pruned channel indices, etc. (experimental).

    +
    +

    xquant

    + +

    exporter

    diff --git a/docs/api/api_docs/methods/get_keras_data_generation_config.html b/docs/api/api_docs/methods/get_keras_data_generation_config.html index 2ab034da0..acacb952d 100644 --- a/docs/api/api_docs/methods/get_keras_data_generation_config.html +++ b/docs/api/api_docs/methods/get_keras_data_generation_config.html @@ -7,7 +7,7 @@ - Get DataGenerationConfig for Keras Models — MCT Documentation: ver 2.1.0 + Get DataGenerationConfig for Keras Models — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Get DataGenerationConfig for Keras Models
    • @@ -45,7 +45,7 @@

    Navigation

    Get DataGenerationConfig for Keras Models

    -model_compression_toolkit.data_generation.get_keras_data_generation_config(n_iter=DEFAULT_N_ITER, optimizer=Adam, data_gen_batch_size=DEFAULT_DATA_GEN_BS, initial_lr=DEFAULT_KERAS_INITIAL_LR, output_loss_multiplier=DEFAULT_KERAS_OUTPUT_LOSS_MULTIPLIER, scheduler_type=SchedulerType.REDUCE_ON_PLATEAU, bn_alignment_loss_type=BatchNormAlignemntLossType.L2_SQUARE, output_loss_type=OutputLossType.REGULARIZED_MIN_MAX_DIFF, data_init_type=DataInitType.Gaussian, layer_weighting_type=BNLayerWeightingType.AVERAGE, image_granularity=ImageGranularity.BatchWise, image_pipeline_type=ImagePipelineType.RANDOM_CROP_FLIP, image_normalization_type=ImageNormalizationType.KERAS_APPLICATIONS, extra_pixels=0, bn_layer_types=[BatchNormalization], clip_images=True, reflection=True)
    +model_compression_toolkit.data_generation.get_keras_data_generation_config(n_iter=DEFAULT_N_ITER, optimizer=Adam, data_gen_batch_size=DEFAULT_DATA_GEN_BS, initial_lr=DEFAULT_KERAS_INITIAL_LR, output_loss_multiplier=DEFAULT_KERAS_OUTPUT_LOSS_MULTIPLIER, scheduler_type=SchedulerType.REDUCE_ON_PLATEAU, bn_alignment_loss_type=BatchNormAlignemntLossType.L2_SQUARE, output_loss_type=OutputLossType.REGULARIZED_MIN_MAX_DIFF, data_init_type=DataInitType.Gaussian, layer_weighting_type=BNLayerWeightingType.AVERAGE, image_granularity=ImageGranularity.BatchWise, image_pipeline_type=ImagePipelineType.SMOOTHING_AND_AUGMENTATION, image_normalization_type=ImageNormalizationType.KERAS_APPLICATIONS, extra_pixels=DEFAULT_KERAS_EXTRA_PIXELS, bn_layer_types=[BatchNormalization], image_clipping=False)

    Function to create a DataGenerationConfig object with the specified configuration parameters.

    Parameters:
    @@ -63,10 +63,9 @@

    Navigation

  • image_granularity (ImageGranularity) – The granularity of the images for optimization.

  • image_pipeline_type (ImagePipelineType) – The type of image pipeline to use.

  • image_normalization_type (ImageNormalizationType) – The type of image normalization to use.

    • -

  • extra_pixels (int) – Extra pixels to add to the input image size. Defaults to 0.

    • +

  • extra_pixels (Union[int, Tuple[int, int]]) – Extra pixels to add to the input image size. Defaults to 0.

  • bn_layer_types (List) – List of BatchNorm layer types to be considered for data generation.

    • -

  • clip_images (bool) – Whether to clip images during optimization.

    • -

  • reflection (bool) – Whether to use reflection during optimization.

    • +

  • image_clipping (bool) – Whether to clip images during optimization.

    • Returns:
    @@ -110,7 +109,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Get DataGenerationConfig for Keras Models
    • diff --git a/docs/api/api_docs/methods/get_keras_gptq_config.html b/docs/api/api_docs/methods/get_keras_gptq_config.html index 808f5f654..37cbf9274 100644 --- a/docs/api/api_docs/methods/get_keras_gptq_config.html +++ b/docs/api/api_docs/methods/get_keras_gptq_config.html @@ -7,7 +7,7 @@ - Get GradientPTQConfig for Keras Models — MCT Documentation: ver 2.1.0 + Get GradientPTQConfig for Keras Models — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Get GradientPTQConfig for Keras Models
    • @@ -45,7 +45,7 @@

    Navigation

    Get GradientPTQConfig for Keras Models

    -model_compression_toolkit.gptq.get_keras_gptq_config(n_epochs, optimizer=tf.keras.optimizers.Adam(learning_rate=LR_DEFAULT), optimizer_rest=tf.keras.optimizers.Adam(learning_rate=LR_REST_DEFAULT), loss=GPTQMultipleTensorsLoss(), log_function=None, use_hessian_based_weights=True, regularization_factor=REG_DEFAULT)
    +model_compression_toolkit.gptq.get_keras_gptq_config(n_epochs, optimizer=tf.keras.optimizers.Adam(learning_rate=LR_DEFAULT), optimizer_rest=tf.keras.optimizers.Adam(learning_rate=LR_REST_DEFAULT), loss=GPTQMultipleTensorsLoss(), log_function=None, use_hessian_based_weights=True, regularization_factor=REG_DEFAULT, hessian_batch_size=ACT_HESSIAN_DEFAULT_BATCH_SIZE)

    Create a GradientPTQConfigV2 instance for Keras models.

    Parameters:
    @@ -57,6 +57,7 @@

    Navigation

  • log_function (Callable) – Function to log information about the gptq process.

  • use_hessian_based_weights (bool) – Whether to use Hessian-based weights for weighted average loss.

  • regularization_factor (float) – A floating point number that defines the regularization factor.

    • +

  • hessian_batch_size (int) – Batch size for Hessian computation in Hessian-based weights GPTQ.

    • Returns:
    @@ -114,7 +115,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Get GradientPTQConfig for Keras Models
    • diff --git a/docs/api/api_docs/methods/get_pytorch_data_generation_config.html b/docs/api/api_docs/methods/get_pytorch_data_generation_config.html index 5d438564e..0edeab3fb 100644 --- a/docs/api/api_docs/methods/get_pytorch_data_generation_config.html +++ b/docs/api/api_docs/methods/get_pytorch_data_generation_config.html @@ -7,7 +7,7 @@ - Get DataGenerationConfig for Pytorch Models — MCT Documentation: ver 2.1.0 + Get DataGenerationConfig for Pytorch Models — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Get DataGenerationConfig for Pytorch Models
    • @@ -45,7 +45,7 @@

    Navigation

    Get DataGenerationConfig for Pytorch Models

    -model_compression_toolkit.data_generation.get_pytorch_data_generation_config(n_iter=DEFAULT_N_ITER, optimizer=RAdam, data_gen_batch_size=DEFAULT_DATA_GEN_BS, initial_lr=DEFAULT_PYTORCH_INITIAL_LR, output_loss_multiplier=DEFAULT_PYTORCH_OUTPUT_LOSS_MULTIPLIER, scheduler_type=SchedulerType.REDUCE_ON_PLATEAU, bn_alignment_loss_type=BatchNormAlignemntLossType.L2_SQUARE, output_loss_type=OutputLossType.REGULARIZED_MIN_MAX_DIFF, data_init_type=DataInitType.Diverse, layer_weighting_type=BNLayerWeightingType.AVERAGE, image_granularity=ImageGranularity.AllImages, image_pipeline_type=ImagePipelineType.RANDOM_CROP, image_normalization_type=ImageNormalizationType.TORCHVISION, extra_pixels=0, bn_layer_types=DEFAULT_PYTORCH_BN_LAYER_TYPES, last_layer_types=DEFAULT_PYTORCH_LAST_LAYER_TYPES, clip_images=True, reflection=True)
    +model_compression_toolkit.data_generation.get_pytorch_data_generation_config(n_iter=DEFAULT_N_ITER, optimizer=RAdam, data_gen_batch_size=DEFAULT_DATA_GEN_BS, initial_lr=DEFAULT_PYTORCH_INITIAL_LR, output_loss_multiplier=DEFAULT_PYTORCH_OUTPUT_LOSS_MULTIPLIER, scheduler_type=SchedulerType.REDUCE_ON_PLATEAU_WITH_RESET, bn_alignment_loss_type=BatchNormAlignemntLossType.L2_SQUARE, output_loss_type=OutputLossType.NEGATIVE_MIN_MAX_DIFF, data_init_type=DataInitType.Gaussian, layer_weighting_type=BNLayerWeightingType.AVERAGE, image_granularity=ImageGranularity.AllImages, image_pipeline_type=ImagePipelineType.SMOOTHING_AND_AUGMENTATION, image_normalization_type=ImageNormalizationType.TORCHVISION, extra_pixels=DEFAULT_PYTORCH_EXTRA_PIXELS, bn_layer_types=DEFAULT_PYTORCH_BN_LAYER_TYPES, last_layer_types=DEFAULT_PYTORCH_LAST_LAYER_TYPES, image_clipping=True)

    Function to create a DataGenerationConfig object with the specified configuration parameters.

    Parameters:
    @@ -63,11 +63,10 @@

    Navigation

  • image_granularity (ImageGranularity) – The granularity of the images for optimization.

  • image_pipeline_type (ImagePipelineType) – The type of image pipeline to use.

  • image_normalization_type (ImageNormalizationType) – The type of image normalization to use.

    • -

  • extra_pixels (int) – Extra pixels to add to the input image size. Defaults to 0.

    • +

  • extra_pixels (Union[int, Tuple[int, int]]) – Extra pixels to add to the input image size. Defaults to 0.

  • bn_layer_types (List) – List of BatchNorm layer types to be considered for data generation.

  • last_layer_types (List) – List of layer types to be considered for the output loss.

    • -

  • clip_images (bool) – Whether to clip images during optimization.

    • -

  • reflection (bool) – Whether to use reflection during optimization.

    • +

  • image_clipping (bool) – Whether to clip images during optimization.

    • Returns:
    @@ -111,7 +110,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Get DataGenerationConfig for Pytorch Models
    • diff --git a/docs/api/api_docs/methods/get_pytroch_gptq_config.html b/docs/api/api_docs/methods/get_pytroch_gptq_config.html index 18485d70a..72de25f28 100644 --- a/docs/api/api_docs/methods/get_pytroch_gptq_config.html +++ b/docs/api/api_docs/methods/get_pytroch_gptq_config.html @@ -7,7 +7,7 @@ - Get GradientPTQConfig for Pytorch Models — MCT Documentation: ver 2.1.0 + Get GradientPTQConfig for Pytorch Models — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Get GradientPTQConfig for Pytorch Models
    • @@ -45,7 +45,7 @@

    Navigation

    Get GradientPTQConfig for Pytorch Models

    -model_compression_toolkit.gptq.get_pytorch_gptq_config(n_epochs, optimizer=Adam([torch.Tensor([])], lr=LR_DEFAULT), optimizer_rest=Adam([torch.Tensor([])], lr=LR_REST_DEFAULT), loss=multiple_tensors_mse_loss, log_function=None, use_hessian_based_weights=True, regularization_factor=REG_DEFAULT)
    +model_compression_toolkit.gptq.get_pytorch_gptq_config(n_epochs, optimizer=Adam([torch.Tensor([])], lr=LR_DEFAULT), optimizer_rest=Adam([torch.Tensor([])], lr=LR_REST_DEFAULT), loss=multiple_tensors_mse_loss, log_function=None, use_hessian_based_weights=True, regularization_factor=REG_DEFAULT, hessian_batch_size=ACT_HESSIAN_DEFAULT_BATCH_SIZE)

    Create a GradientPTQConfigV2 instance for Pytorch models.

    Parameters:
    @@ -57,6 +57,7 @@

    Navigation

  • log_function (Callable) – Function to log information about the gptq process.

  • use_hessian_based_weights (bool) – Whether to use Hessian-based weights for weighted average loss.

  • regularization_factor (float) – A floating point number that defines the regularization factor.

    • +

  • hessian_batch_size (int) – Batch size for Hessian computation in Hessian-based weights GPTQ.

    • Returns:
    @@ -111,7 +112,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Get GradientPTQConfig for Pytorch Models
    • diff --git a/docs/api/api_docs/methods/get_target_platform_capabilities.html b/docs/api/api_docs/methods/get_target_platform_capabilities.html index de98bbca6..5fe03a901 100644 --- a/docs/api/api_docs/methods/get_target_platform_capabilities.html +++ b/docs/api/api_docs/methods/get_target_platform_capabilities.html @@ -7,7 +7,7 @@ - Get TargetPlatformCapabilities — MCT Documentation: ver 2.1.0 + Get TargetPlatformCapabilities — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Get TargetPlatformCapabilities
    • @@ -104,7 +104,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Get TargetPlatformCapabilities
    • diff --git a/docs/api/api_docs/methods/keras_data_generation_experimental.html b/docs/api/api_docs/methods/keras_data_generation_experimental.html index 200907427..3094a6bc0 100644 --- a/docs/api/api_docs/methods/keras_data_generation_experimental.html +++ b/docs/api/api_docs/methods/keras_data_generation_experimental.html @@ -7,7 +7,7 @@ - Keras Data Generation — MCT Documentation: ver 2.1.0 + Keras Data Generation — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Keras Data Generation
    • @@ -52,7 +52,7 @@

    Navigation

    • model (Model) – Keras model to generate data for.

    • n_images (int) – Number of images to generate.

      • -

    • output_image_size (Tuple) – Size of the output images.

      • +

    • output_image_size (Union[int, Tuple[int, int]]) – Size of the output images.

    • data_generation_config (DataGenerationConfig) – Configuration for data generation.

    @@ -122,7 +122,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Keras Data Generation
    • diff --git a/docs/api/api_docs/methods/keras_gradient_post_training_quantization.html b/docs/api/api_docs/methods/keras_gradient_post_training_quantization.html index 58a8f49bb..62eb33a89 100644 --- a/docs/api/api_docs/methods/keras_gradient_post_training_quantization.html +++ b/docs/api/api_docs/methods/keras_gradient_post_training_quantization.html @@ -7,7 +7,7 @@ - Keras Gradient Based Post Training Quantization — MCT Documentation: ver 2.1.0 + Keras Gradient Based Post Training Quantization — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Keras Gradient Based Post Training Quantization
    • @@ -152,7 +152,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Keras Gradient Based Post Training Quantization
    • diff --git a/docs/api/api_docs/methods/keras_kpi_data.html b/docs/api/api_docs/methods/keras_kpi_data.html index 79b63e7da..e39322e8c 100644 --- a/docs/api/api_docs/methods/keras_kpi_data.html +++ b/docs/api/api_docs/methods/keras_kpi_data.html @@ -7,7 +7,7 @@ - Get Resource Utilization information for Keras Models — MCT Documentation: ver 2.1.0 + Get Resource Utilization information for Keras Models — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Get Resource Utilization information for Keras Models
    • @@ -115,7 +115,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Get Resource Utilization information for Keras Models
    • diff --git a/docs/api/api_docs/methods/keras_load_quantizad_model.html b/docs/api/api_docs/methods/keras_load_quantizad_model.html index 8e23f0b67..db16f181a 100644 --- a/docs/api/api_docs/methods/keras_load_quantizad_model.html +++ b/docs/api/api_docs/methods/keras_load_quantizad_model.html @@ -7,7 +7,7 @@ - Load Quantized Keras Model — MCT Documentation: ver 2.1.0 + Load Quantized Keras Model — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Load Quantized Keras Model
    • @@ -89,7 +89,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Load Quantized Keras Model
    • diff --git a/docs/api/api_docs/methods/keras_post_training_quantization.html b/docs/api/api_docs/methods/keras_post_training_quantization.html index 3bc4d16c1..9cc04bf05 100644 --- a/docs/api/api_docs/methods/keras_post_training_quantization.html +++ b/docs/api/api_docs/methods/keras_post_training_quantization.html @@ -7,7 +7,7 @@ - Keras Post Training Quantization — MCT Documentation: ver 2.1.0 + Keras Post Training Quantization — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Keras Post Training Quantization
    • @@ -144,7 +144,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Keras Post Training Quantization
    • diff --git a/docs/api/api_docs/methods/keras_pruning_experimental.html b/docs/api/api_docs/methods/keras_pruning_experimental.html index 5ca019ea6..0574b4a2e 100644 --- a/docs/api/api_docs/methods/keras_pruning_experimental.html +++ b/docs/api/api_docs/methods/keras_pruning_experimental.html @@ -7,7 +7,7 @@ - Keras Structured Pruning — MCT Documentation: ver 2.1.0 + Keras Structured Pruning — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Keras Structured Pruning
    • @@ -145,7 +145,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Keras Structured Pruning
    • diff --git a/docs/api/api_docs/methods/keras_quantization_aware_training_finalize_experimental.html b/docs/api/api_docs/methods/keras_quantization_aware_training_finalize_experimental.html index 4c78e161e..f19e24b23 100644 --- a/docs/api/api_docs/methods/keras_quantization_aware_training_finalize_experimental.html +++ b/docs/api/api_docs/methods/keras_quantization_aware_training_finalize_experimental.html @@ -7,7 +7,7 @@ - Keras Quantization Aware Training Model Finalize — MCT Documentation: ver 2.1.0 + Keras Quantization Aware Training Model Finalize — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Keras Quantization Aware Training Model Finalize
    • @@ -132,7 +132,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Keras Quantization Aware Training Model Finalize
    • diff --git a/docs/api/api_docs/methods/keras_quantization_aware_training_init_experimental.html b/docs/api/api_docs/methods/keras_quantization_aware_training_init_experimental.html index 138081f36..b5a47c72a 100644 --- a/docs/api/api_docs/methods/keras_quantization_aware_training_init_experimental.html +++ b/docs/api/api_docs/methods/keras_quantization_aware_training_init_experimental.html @@ -7,7 +7,7 @@ - Keras Quantization Aware Training Model Init — MCT Documentation: ver 2.1.0 + Keras Quantization Aware Training Model Init — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Keras Quantization Aware Training Model Init
    • @@ -152,7 +152,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Keras Quantization Aware Training Model Init
    • diff --git a/docs/api/api_docs/methods/pytorch_data_generation_experimental.html b/docs/api/api_docs/methods/pytorch_data_generation_experimental.html index 15b0a822a..14fc9f215 100644 --- a/docs/api/api_docs/methods/pytorch_data_generation_experimental.html +++ b/docs/api/api_docs/methods/pytorch_data_generation_experimental.html @@ -7,7 +7,7 @@ - Pytorch Data Generation — MCT Documentation: ver 2.1.0 + Pytorch Data Generation — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Pytorch Data Generation
    • @@ -52,7 +52,7 @@

    Navigation

    • model (Module) – PyTorch model to generate data for.

    • n_images (int) – Number of images to generate.

      • -

    • output_image_size (int) – The hight and width size of the output images.

      • +

    • output_image_size (Union[int, Tuple[int, int]]) – The hight and width size of the output images.

    • data_generation_config (DataGenerationConfig) – Configuration for data generation.

    @@ -122,7 +122,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Pytorch Data Generation
    • diff --git a/docs/api/api_docs/methods/pytorch_gradient_post_training_quantization.html b/docs/api/api_docs/methods/pytorch_gradient_post_training_quantization.html index d1767497b..3db8f08ed 100644 --- a/docs/api/api_docs/methods/pytorch_gradient_post_training_quantization.html +++ b/docs/api/api_docs/methods/pytorch_gradient_post_training_quantization.html @@ -7,7 +7,7 @@ - Pytorch Gradient Based Post Training Quantization — MCT Documentation: ver 2.1.0 + Pytorch Gradient Based Post Training Quantization — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Pytorch Gradient Based Post Training Quantization
    • @@ -134,7 +134,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Pytorch Gradient Based Post Training Quantization
    • diff --git a/docs/api/api_docs/methods/pytorch_kpi_data.html b/docs/api/api_docs/methods/pytorch_kpi_data.html index 83c3caf20..6913f0e11 100644 --- a/docs/api/api_docs/methods/pytorch_kpi_data.html +++ b/docs/api/api_docs/methods/pytorch_kpi_data.html @@ -7,7 +7,7 @@ - Get Resource Utilization information for PyTorch Models — MCT Documentation: ver 2.1.0 + Get Resource Utilization information for PyTorch Models — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Get Resource Utilization information for PyTorch Models
    • @@ -113,7 +113,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Get Resource Utilization information for PyTorch Models
    • diff --git a/docs/api/api_docs/methods/pytorch_post_training_quantization.html b/docs/api/api_docs/methods/pytorch_post_training_quantization.html index 375535673..d69365f83 100644 --- a/docs/api/api_docs/methods/pytorch_post_training_quantization.html +++ b/docs/api/api_docs/methods/pytorch_post_training_quantization.html @@ -7,7 +7,7 @@ - Pytorch Post Training Quantization — MCT Documentation: ver 2.1.0 + Pytorch Post Training Quantization — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Pytorch Post Training Quantization
    • @@ -123,7 +123,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Pytorch Post Training Quantization
    • diff --git a/docs/api/api_docs/methods/pytorch_pruning_experimental.html b/docs/api/api_docs/methods/pytorch_pruning_experimental.html index aed9398c6..c2eee01ab 100644 --- a/docs/api/api_docs/methods/pytorch_pruning_experimental.html +++ b/docs/api/api_docs/methods/pytorch_pruning_experimental.html @@ -7,7 +7,7 @@ - Pytorch Structured Pruning — MCT Documentation: ver 2.1.0 + Pytorch Structured Pruning — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Pytorch Structured Pruning
    • @@ -146,7 +146,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Pytorch Structured Pruning
    • diff --git a/docs/api/api_docs/methods/pytorch_quantization_aware_training_finalize_experimental.html b/docs/api/api_docs/methods/pytorch_quantization_aware_training_finalize_experimental.html index 9eb722711..7fb23e0f1 100644 --- a/docs/api/api_docs/methods/pytorch_quantization_aware_training_finalize_experimental.html +++ b/docs/api/api_docs/methods/pytorch_quantization_aware_training_finalize_experimental.html @@ -7,7 +7,7 @@ - PyTorch Quantization Aware Training Model Finalize — MCT Documentation: ver 2.1.0 + PyTorch Quantization Aware Training Model Finalize — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • PyTorch Quantization Aware Training Model Finalize
    • @@ -115,7 +115,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • PyTorch Quantization Aware Training Model Finalize
    • diff --git a/docs/api/api_docs/methods/pytorch_quantization_aware_training_init_experimental.html b/docs/api/api_docs/methods/pytorch_quantization_aware_training_init_experimental.html index 10180bc2b..2263184a6 100644 --- a/docs/api/api_docs/methods/pytorch_quantization_aware_training_init_experimental.html +++ b/docs/api/api_docs/methods/pytorch_quantization_aware_training_init_experimental.html @@ -7,7 +7,7 @@ - PyTorch Quantization Aware Training Model Init — MCT Documentation: ver 2.1.0 + PyTorch Quantization Aware Training Model Init — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • PyTorch Quantization Aware Training Model Init
    • @@ -139,7 +139,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • PyTorch Quantization Aware Training Model Init
    • diff --git a/docs/api/api_docs/methods/set_logger_path.html b/docs/api/api_docs/methods/set_logger_path.html index 4914d0afc..46bc672ee 100644 --- a/docs/api/api_docs/methods/set_logger_path.html +++ b/docs/api/api_docs/methods/set_logger_path.html @@ -7,7 +7,7 @@ - Enable a Logger — MCT Documentation: ver 2.1.0 + Enable a Logger — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Enable a Logger
    • @@ -86,7 +86,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Enable a Logger
    • diff --git a/docs/api/api_docs/methods/xquant_report_keras_experimental.html b/docs/api/api_docs/methods/xquant_report_keras_experimental.html new file mode 100644 index 000000000..6a019e06a --- /dev/null +++ b/docs/api/api_docs/methods/xquant_report_keras_experimental.html @@ -0,0 +1,110 @@ + + + + + + + + + + XQuant Report Keras — MCT Documentation: ver 2.2.0 + + + + + + + + + + + + + + + + + +
    +
    +
    +
    + +
    +

    XQuant Report Keras

    +
    +
    +model_compression_toolkit.xquant.keras.facade_xquant_report.xquant_report_keras_experimental(float_model, quantized_model, repr_dataset, validation_dataset, xquant_config)
    +

    Generate an explainable quantization report for a quantized Keras model.

    +
    +
    Parameters:
    +
      +

    • float_model (keras.Model) – The original floating-point Keras model.

      • +

    • quantized_model (keras.Model) – The quantized Keras model.

      • +

    • repr_dataset (Callable) – The representative dataset used during quantization for similarity metrics computation.

      • +

    • validation_dataset (Callable) – The validation dataset used for evaluation for similarity metrics computation.

      • +

    • xquant_config (XQuantConfig) – Configuration settings for explainable quantization.

      • +
    +
    +
    Returns:
    +

    A dictionary containing the collected similarity metrics and report data.

    +
    +
    Return type:
    +

    Dict[str, Any]

    +
    +
    Return type:
    +

    Dict[str, Any]

    +
    +
    +
    + +
    + + +
    +
    +
    +
    + +
    +
    + + + + \ No newline at end of file diff --git a/docs/api/api_docs/methods/xquant_report_pytorch_experimental.html b/docs/api/api_docs/methods/xquant_report_pytorch_experimental.html new file mode 100644 index 000000000..cf0cb52c4 --- /dev/null +++ b/docs/api/api_docs/methods/xquant_report_pytorch_experimental.html @@ -0,0 +1,107 @@ + + + + + + + + + + XQuant Report Pytorch — MCT Documentation: ver 2.2.0 + + + + + + + + + + + + + + + + + +
    +
    +
    +
    + +
    +

    XQuant Report Pytorch

    +
    +
    +model_compression_toolkit.xquant.pytorch.facade_xquant_report.xquant_report_pytorch_experimental(float_model, quantized_model, repr_dataset, validation_dataset, xquant_config)
    +

    Generate an explainable quantization report for a quantized Pytorch model.

    +
    +
    Parameters:
    +
      +

    • float_model (torch.nn.Module) – The original floating-point Pytorch model.

      • +

    • quantized_model (torch.nn.Module) – The quantized Pytorch model.

      • +

    • repr_dataset (Callable) – The representative dataset used during quantization.

      • +

    • validation_dataset (Callable) – The validation dataset used for evaluation.

      • +

    • xquant_config (XQuantConfig) – Configuration settings for explainable quantization.

      • +
    +
    +
    Returns:
    +

    A dictionary containing the collected similarity metrics and report data.

    +
    +
    Return type:
    +

    Dict[str, Any]

    +
    +
    +
    + +
    + + +
    +
    +
    +
    + +
    +
    + + + + \ No newline at end of file diff --git a/docs/api/api_docs/modules/core_config.html b/docs/api/api_docs/modules/core_config.html index c5d5f80cf..1560526f9 100644 --- a/docs/api/api_docs/modules/core_config.html +++ b/docs/api/api_docs/modules/core_config.html @@ -7,7 +7,7 @@ - CoreConfig — MCT Documentation: ver 2.1.0 + CoreConfig — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • CoreConfig
    • @@ -46,7 +46,7 @@

    Navigation

    Class to configure the optimization process of the model:

    -class model_compression_toolkit.core.CoreConfig(quantization_config=QuantizationConfig(), mixed_precision_config=None, debug_config=DebugConfig())
    +class model_compression_toolkit.core.CoreConfig(quantization_config=None, mixed_precision_config=None, bit_width_config=None, debug_config=None)

    A class to hold the configurations classes of the MCT-core.

    Parameters:
    @@ -55,6 +55,7 @@

    Navigation

  • mixed_precision_config (MixedPrecisionQuantizationConfig) – Config for mixed precision quantization.

  • None (If) –

  • used. (a default MixedPrecisionQuantizationConfig is) –

    • +

  • bit_width_config (BitWidthConfig) – Config for manual bit-width selection.

  • debug_config (DebugConfig) – Config for debugging and editing the network quantization process.

    • @@ -90,7 +91,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • CoreConfig
    • diff --git a/docs/api/api_docs/modules/debug_config.html b/docs/api/api_docs/modules/debug_config.html index 7eb7dcf98..faf43f328 100644 --- a/docs/api/api_docs/modules/debug_config.html +++ b/docs/api/api_docs/modules/debug_config.html @@ -7,7 +7,7 @@ - debug_config Module — MCT Documentation: ver 2.1.0 + debug_config Module — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • debug_config Module
    • @@ -48,7 +48,7 @@

    DebugConfig
    -class model_compression_toolkit.core.DebugConfig(analyze_similarity=False, network_editor=[])
    +class model_compression_toolkit.core.DebugConfig(analyze_similarity=False, network_editor=[], simulate_scheduler=False)

    A class for MCT core debug information.

    Parameters:
    @@ -56,6 +56,7 @@

    DebugConfigEditRule]) – A list of rules and actions to edit the network for quantization.

    +

  • simulate_scheduler (bool) – Simulate scheduler behaviour to compute operators order and cuts.

    • @@ -101,7 +102,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • debug_config Module
    • diff --git a/docs/api/api_docs/modules/exporter.html b/docs/api/api_docs/modules/exporter.html index 745a849c7..8c98c0976 100644 --- a/docs/api/api_docs/modules/exporter.html +++ b/docs/api/api_docs/modules/exporter.html @@ -7,7 +7,7 @@ - exporter Module — MCT Documentation: ver 2.1.0 + exporter Module — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • exporter Module
    • @@ -318,7 +318,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • exporter Module
    • diff --git a/docs/api/api_docs/modules/layer_filters.html b/docs/api/api_docs/modules/layer_filters.html index f2e31f98e..659abc0f2 100644 --- a/docs/api/api_docs/modules/layer_filters.html +++ b/docs/api/api_docs/modules/layer_filters.html @@ -7,7 +7,7 @@ - Layer Attributes Filters — MCT Documentation: ver 2.1.0 + Layer Attributes Filters — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Layer Attributes Filters
    • @@ -214,7 +214,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Layer Attributes Filters
    • diff --git a/docs/api/api_docs/modules/network_editor.html b/docs/api/api_docs/modules/network_editor.html index fa694b250..4d12299f3 100644 --- a/docs/api/api_docs/modules/network_editor.html +++ b/docs/api/api_docs/modules/network_editor.html @@ -7,7 +7,7 @@ - network_editor Module — MCT Documentation: ver 2.1.0 + network_editor Module — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • network_editor Module
    • @@ -283,7 +283,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • network_editor Module
    • diff --git a/docs/api/api_docs/modules/qat_config.html b/docs/api/api_docs/modules/qat_config.html index b2013ad11..787a31493 100644 --- a/docs/api/api_docs/modules/qat_config.html +++ b/docs/api/api_docs/modules/qat_config.html @@ -7,7 +7,7 @@ - qat_config Module — MCT Documentation: ver 2.1.0 + qat_config Module — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • qat_config Module
    • @@ -119,7 +119,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • qat_config Module
    • diff --git a/docs/api/api_docs/modules/target_platform.html b/docs/api/api_docs/modules/target_platform.html index 7b71d9ed8..4f799584a 100644 --- a/docs/api/api_docs/modules/target_platform.html +++ b/docs/api/api_docs/modules/target_platform.html @@ -7,7 +7,7 @@ - target_platform Module — MCT Documentation: ver 2.1.0 + target_platform Module — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • target_platform Module
    • @@ -88,7 +88,7 @@

    QuantizationMethod

    -class model_compression_toolkit.target_platform.OpQuantizationConfig(default_weight_attr_config, attr_weights_configs_mapping, activation_quantization_method, activation_n_bits, enable_activation_quantization, quantization_preserving, fixed_scale, fixed_zero_point, simd_size)
    +class model_compression_toolkit.target_platform.OpQuantizationConfig(default_weight_attr_config, attr_weights_configs_mapping, activation_quantization_method, activation_n_bits, supported_input_activation_n_bits, enable_activation_quantization, quantization_preserving, fixed_scale, fixed_zero_point, simd_size, signedness)

    OpQuantizationConfig is a class to configure the quantization parameters of an operator.

    Parameters:
    @@ -97,11 +97,13 @@

    OpQuantizationConfig

    attr_weights_configs_mapping (Dict[str, AttributeQuantizationConfig]) – A mapping between an op attribute name and its quantization configuration.

  • activation_quantization_method (QuantizationMethod) – Which method to use from QuantizationMethod for activation quantization.

  • activation_n_bits (int) – Number of bits to quantize the activations.

    • +

  • supported_input_activation_n_bits (int or Tuple[int]) – Number of bits that operator accepts as input.

  • enable_activation_quantization (bool) – Whether to quantize the model activations or not.

  • quantization_preserving (bool) – Whether quantization parameters should be the same for an operator’s input and output.

  • fixed_scale (float) – Scale to use for an operator quantization parameters.

  • fixed_zero_point (int) – Zero-point to use for an operator quantization parameters.

  • simd_size (int) – Per op integer representing the Single Instruction, Multiple Data (SIMD) width of an operator. It indicates the number of data elements that can be fetched and processed simultaneously in a single instruction.

    • +

  • signedness (bool) – Set activation quantization signedness.

    • @@ -336,7 +338,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • target_platform Module
    • diff --git a/docs/api/api_docs/modules/trainable_infrastructure.html b/docs/api/api_docs/modules/trainable_infrastructure.html index bde7de478..27074f6e2 100644 --- a/docs/api/api_docs/modules/trainable_infrastructure.html +++ b/docs/api/api_docs/modules/trainable_infrastructure.html @@ -7,7 +7,7 @@ - trainable_infrastructure Module — MCT Documentation: ver 2.1.0 + trainable_infrastructure Module — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • trainable_infrastructure Module
    • @@ -204,7 +204,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • trainable_infrastructure Module
    • diff --git a/docs/api/api_docs/notes/tpc_note.html b/docs/api/api_docs/notes/tpc_note.html index dcf330b6c..44c10f5d3 100644 --- a/docs/api/api_docs/notes/tpc_note.html +++ b/docs/api/api_docs/notes/tpc_note.html @@ -7,7 +7,7 @@ - <no title> — MCT Documentation: ver 2.1.0 + <no title> — MCT Documentation: ver 2.2.0 @@ -31,7 +31,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • <no title>
    • @@ -77,7 +77,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • <no title>
    • diff --git a/docs/genindex.html b/docs/genindex.html index 4396625fc..4382c5f92 100644 --- a/docs/genindex.html +++ b/docs/genindex.html @@ -6,7 +6,7 @@ - Index — MCT Documentation: ver 2.1.0 + Index — MCT Documentation: ver 2.2.0 @@ -30,7 +30,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Index
    • @@ -62,6 +62,7 @@

    Index

    | R | S | T + | X

    A

    @@ -82,10 +83,14 @@

    B

  • BaseKerasTrainableQuantizer (class in model_compression_toolkit.trainable_infrastructure)
  • BasePytorchTrainableQuantizer (class in model_compression_toolkit.trainable_infrastructure) +
    • +
  • BatchNormAlignemntLossType (class in model_compression_toolkit.data_generation)
    • diff --git a/docs/guidelines/visualization.html b/docs/guidelines/visualization.html index 83ba2128f..7d43406fc 100644 --- a/docs/guidelines/visualization.html +++ b/docs/guidelines/visualization.html @@ -7,7 +7,7 @@ - Visualization within TensorBoard — MCT Documentation: ver 2.1.0 + Visualization within TensorBoard — MCT Documentation: ver 2.2.0 @@ -39,7 +39,7 @@

    Navigation

  • previous |
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Visualization within TensorBoard
    • @@ -186,7 +186,7 @@

    Navigation

  • previous |
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Visualization within TensorBoard
    • diff --git a/docs/index.html b/docs/index.html index 4bce16c28..0c9e7fd19 100644 --- a/docs/index.html +++ b/docs/index.html @@ -7,7 +7,7 @@ - Model Compression Toolkit User Guide — MCT Documentation: ver 2.1.0 + Model Compression Toolkit User Guide — MCT Documentation: ver 2.2.0 @@ -35,7 +35,7 @@

    Navigation

  • next |
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Model Compression Toolkit User Guide
    • @@ -181,7 +181,7 @@

    Navigation

  • next |
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Model Compression Toolkit User Guide
    • diff --git a/docs/objects.inv b/docs/objects.inv index 404a53be2..01caa1317 100644 Binary files a/docs/objects.inv and b/docs/objects.inv differ diff --git a/docs/search.html b/docs/search.html index b7d975479..25f88fb35 100644 --- a/docs/search.html +++ b/docs/search.html @@ -6,7 +6,7 @@ - Search — MCT Documentation: ver 2.1.0 + Search — MCT Documentation: ver 2.2.0 @@ -36,7 +36,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Search
    • @@ -93,7 +93,7 @@

    Navigation

  • index
    • -
  • MCT Documentation: ver 2.1.0 »
    • +
  • MCT Documentation: ver 2.2.0 »
  • Search
    • diff --git a/docs/searchindex.js b/docs/searchindex.js index 5db871543..6bb933530 100644 --- a/docs/searchindex.js +++ b/docs/searchindex.js @@ -1 +1 @@ -Search.setIndex({"docnames": ["api/api_docs/classes/DataGenerationConfig", "api/api_docs/classes/DefaultDict", "api/api_docs/classes/FrameworkInfo", "api/api_docs/classes/GradientPTQConfig", "api/api_docs/classes/MixedPrecisionQuantizationConfig", "api/api_docs/classes/MpDistanceWeighting", "api/api_docs/classes/PruningConfig", "api/api_docs/classes/PruningInfo", "api/api_docs/classes/QuantizationConfig", "api/api_docs/classes/QuantizationErrorMethod", "api/api_docs/classes/ResourceUtilization", "api/api_docs/index", "api/api_docs/methods/get_keras_data_generation_config", "api/api_docs/methods/get_keras_gptq_config", "api/api_docs/methods/get_pytorch_data_generation_config", "api/api_docs/methods/get_pytroch_gptq_config", "api/api_docs/methods/get_target_platform_capabilities", "api/api_docs/methods/keras_data_generation_experimental", "api/api_docs/methods/keras_gradient_post_training_quantization", "api/api_docs/methods/keras_kpi_data", "api/api_docs/methods/keras_load_quantizad_model", "api/api_docs/methods/keras_post_training_quantization", "api/api_docs/methods/keras_pruning_experimental", "api/api_docs/methods/keras_quantization_aware_training_finalize_experimental", "api/api_docs/methods/keras_quantization_aware_training_init_experimental", "api/api_docs/methods/pytorch_data_generation_experimental", "api/api_docs/methods/pytorch_gradient_post_training_quantization", "api/api_docs/methods/pytorch_kpi_data", "api/api_docs/methods/pytorch_post_training_quantization", "api/api_docs/methods/pytorch_pruning_experimental", "api/api_docs/methods/pytorch_quantization_aware_training_finalize_experimental", "api/api_docs/methods/pytorch_quantization_aware_training_init_experimental", "api/api_docs/methods/set_logger_path", "api/api_docs/modules/core_config", "api/api_docs/modules/debug_config", "api/api_docs/modules/exporter", "api/api_docs/modules/layer_filters", "api/api_docs/modules/network_editor", "api/api_docs/modules/qat_config", "api/api_docs/modules/target_platform", "api/api_docs/modules/trainable_infrastructure", "api/api_docs/notes/tpc_note", "guidelines/visualization", "index"], "filenames": ["api/api_docs/classes/DataGenerationConfig.rst", "api/api_docs/classes/DefaultDict.rst", "api/api_docs/classes/FrameworkInfo.rst", "api/api_docs/classes/GradientPTQConfig.rst", "api/api_docs/classes/MixedPrecisionQuantizationConfig.rst", "api/api_docs/classes/MpDistanceWeighting.rst", "api/api_docs/classes/PruningConfig.rst", "api/api_docs/classes/PruningInfo.rst", "api/api_docs/classes/QuantizationConfig.rst", "api/api_docs/classes/QuantizationErrorMethod.rst", "api/api_docs/classes/ResourceUtilization.rst", "api/api_docs/index.rst", "api/api_docs/methods/get_keras_data_generation_config.rst", "api/api_docs/methods/get_keras_gptq_config.rst", "api/api_docs/methods/get_pytorch_data_generation_config.rst", "api/api_docs/methods/get_pytroch_gptq_config.rst", "api/api_docs/methods/get_target_platform_capabilities.rst", "api/api_docs/methods/keras_data_generation_experimental.rst", "api/api_docs/methods/keras_gradient_post_training_quantization.rst", "api/api_docs/methods/keras_kpi_data.rst", "api/api_docs/methods/keras_load_quantizad_model.rst", "api/api_docs/methods/keras_post_training_quantization.rst", "api/api_docs/methods/keras_pruning_experimental.rst", "api/api_docs/methods/keras_quantization_aware_training_finalize_experimental.rst", "api/api_docs/methods/keras_quantization_aware_training_init_experimental.rst", "api/api_docs/methods/pytorch_data_generation_experimental.rst", "api/api_docs/methods/pytorch_gradient_post_training_quantization.rst", "api/api_docs/methods/pytorch_kpi_data.rst", "api/api_docs/methods/pytorch_post_training_quantization.rst", "api/api_docs/methods/pytorch_pruning_experimental.rst", "api/api_docs/methods/pytorch_quantization_aware_training_finalize_experimental.rst", "api/api_docs/methods/pytorch_quantization_aware_training_init_experimental.rst", "api/api_docs/methods/set_logger_path.rst", "api/api_docs/modules/core_config.rst", "api/api_docs/modules/debug_config.rst", "api/api_docs/modules/exporter.rst", "api/api_docs/modules/layer_filters.rst", "api/api_docs/modules/network_editor.rst", "api/api_docs/modules/qat_config.rst", "api/api_docs/modules/target_platform.rst", "api/api_docs/modules/trainable_infrastructure.rst", "api/api_docs/notes/tpc_note.rst", "guidelines/visualization.rst", "index.rst"], "titles": ["Data Generation Configuration", "DefaultDict Class", "FrameworkInfo Class", "GradientPTQConfig Class", "MixedPrecisionQuantizationConfig", "MpDistanceWeighting", "Pruning Configuration", "Pruning Information", "QuantizationConfig", "QuantizationErrorMethod", "ResourceUtilization", "API Docs", "Get DataGenerationConfig for Keras Models", "Get GradientPTQConfig for Keras Models", "Get DataGenerationConfig for Pytorch Models", "Get GradientPTQConfig for Pytorch Models", "Get TargetPlatformCapabilities", "Keras Data Generation", "Keras Gradient Based Post Training Quantization", "Get Resource Utilization information for Keras Models", "Load Quantized Keras Model", "Keras Post Training Quantization", "Keras Structured Pruning", "Keras Quantization Aware Training Model Finalize", "Keras Quantization Aware Training Model Init", "Pytorch Data Generation", "Pytorch Gradient Based Post Training Quantization", "Get Resource Utilization information for PyTorch Models", "Pytorch Post Training Quantization", "Pytorch Structured Pruning", "PyTorch Quantization Aware Training Model Finalize", "PyTorch Quantization Aware Training Model Init", "Enable a Logger", "CoreConfig", "debug_config Module", "exporter Module", "Layer Attributes Filters", "network_editor Module", "qat_config Module", "target_platform Module", "trainable_infrastructure Module", "<no title>", "Visualization within TensorBoard", "Model Compression Toolkit User Guide"], "terms": {"class": [0, 4, 5, 6, 7, 8, 9, 10, 11, 20, 33, 34, 35, 36, 37, 38, 39, 40], "model_compression_toolkit": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 42], "data_gener": [0, 12, 14, 17, 25], "datagenerationconfig": [0, 11, 17, 25], "n_iter": [0, 12, 14, 17, 25], "optim": [0, 2, 3, 8, 10, 11, 12, 13, 14, 15, 16, 18, 19, 21, 24, 26, 27, 28, 31, 33, 39, 40, 41, 43], "data_gen_batch_s": [0, 12, 14, 17, 25], "initial_lr": [0, 12, 14], "output_loss_multipli": [0, 12, 14], "image_granular": [0, 12, 14], "allimag": [0, 14], "scheduler_typ": [0, 12, 14], "none": [0, 1, 3, 4, 13, 15, 16, 18, 20, 21, 24, 26, 28, 31, 33, 35, 37, 38, 39, 40], "bn_alignment_loss_typ": [0, 12, 14], "output_loss_typ": [0, 12, 14], "data_init_typ": [0, 12, 14], "layer_weighting_typ": [0, 12, 14], "image_pipeline_typ": [0, 12, 14], "image_normalization_typ": [0, 12, 14], "extra_pixel": [0, 12, 14], "0": [0, 2, 3, 4, 7, 8, 12, 13, 14, 18, 21, 22, 23, 24, 29, 35, 40], "bn_layer_typ": [0, 12, 14], "last_layer_typ": [0, 14], "clip_imag": [0, 12, 14], "true": [0, 3, 4, 8, 12, 13, 14, 15, 20, 30, 31, 40], "reflect": [0, 12, 14], "initi": [0, 1, 3, 6, 12, 14, 24, 31, 39, 40], "paramet": [0, 1, 2, 3, 4, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40], "int": [0, 3, 4, 6, 8, 12, 13, 14, 15, 17, 25, 39, 40], "number": [0, 3, 4, 6, 12, 13, 14, 15, 17, 18, 21, 22, 24, 25, 26, 28, 29, 31, 39, 40], "iter": [0, 12, 14, 17, 18, 21, 24, 25, 26, 28, 31], "ani": [0, 1, 2, 3, 35, 36, 39, 40], "The": [0, 2, 3, 5, 6, 7, 8, 10, 12, 13, 14, 15, 17, 18, 21, 22, 23, 24, 25, 26, 28, 29, 31, 35, 37, 39, 40, 42], "us": [0, 1, 2, 3, 4, 5, 6, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 33, 34, 36, 37, 38, 39, 40, 41, 42, 43], "batch": [0, 12, 14, 17, 18, 21, 24, 25, 26, 28, 31], "size": [0, 12, 14, 17, 18, 21, 23, 24, 25, 31, 35, 38], "float": [0, 3, 4, 8, 12, 13, 14, 15, 18, 24, 26, 28, 31, 35, 39, 40, 42], "learn": [0, 12, 13, 14, 38], "rate": [0, 12, 13, 14], "multipli": [0, 12, 14], "output": [0, 2, 5, 8, 12, 14, 17, 18, 21, 24, 25, 26, 28, 30, 31, 39, 42, 43], "loss": [0, 3, 12, 13, 14, 15, 18, 22, 26, 28, 29], "granular": [0, 12, 14], "imag": [0, 4, 12, 14, 17, 18, 21, 24, 25, 26, 28, 31, 42], "default": [0, 1, 3, 6, 12, 13, 14, 16, 18, 21, 22, 26, 28, 29, 33, 35, 38, 39, 42], "type": [0, 1, 3, 4, 6, 7, 12, 13, 14, 15, 16, 17, 18, 19, 22, 23, 25, 27, 29, 35, 37, 39], "schedul": [0, 12, 14], "batchnorm": [0, 12, 14, 17, 18, 21, 24, 26, 28, 31], "align": [0, 12, 14], "layer": [0, 2, 4, 5, 7, 12, 14, 17, 18, 21, 22, 23, 24, 26, 28, 29, 30, 31, 34, 35, 37, 39, 40, 42], "weight": [0, 2, 3, 4, 5, 8, 10, 11, 12, 13, 14, 15, 18, 19, 22, 24, 26, 27, 28, 29, 30, 31, 35, 37, 38, 39, 40, 42], "pipelin": [0, 12, 14], "normal": [0, 3, 4, 12, 14], "extra": [0, 12, 14], "pixel": [0, 12, 14], "add": [0, 2, 12, 14, 20, 39, 40], "input": [0, 4, 8, 12, 14, 18, 21, 24, 26, 28, 31, 39], "list": [0, 2, 3, 4, 12, 13, 14, 15, 17, 25, 34, 37, 39, 43], "bool": [0, 3, 4, 8, 12, 13, 14, 15, 34, 39, 40], "flag": 0, "enabl": [0, 4, 8, 11, 34, 40, 43], "clip": [0, 12, 14], "valu": [0, 1, 2, 3, 4, 5, 6, 8, 9, 10, 18, 21, 22, 23, 24, 29, 35, 36, 37, 38, 39], "an": [0, 1, 2, 3, 7, 18, 21, 24, 31, 35, 36, 37, 38, 39, 40, 43], "enum": [0, 2, 3, 4, 5, 6, 9, 38], "choos": [0, 3, 35], "depend": [0, 18, 21, 24, 26, 28, 31], "when": [0, 1, 2, 4, 6, 9, 10, 11, 13, 15, 18, 21, 23, 24, 34, 36, 38, 39, 40, 42], "imagewis": 0, "batchwis": [0, 12], "reduce_on_plateau": [0, 12, 14], "reduceonplateau": 0, "step": [0, 38], "l2_squar": [0, 12, 14], "l2": 0, "squar": [0, 9], "No": 0, "appli": [0, 11, 35, 36, 37, 39], "min_max_diff": 0, "min": [0, 2, 8, 9, 18, 21, 24, 26, 28, 31, 42], "max": [0, 2, 8, 9, 18, 19, 21, 24, 26, 27, 28, 31, 42], "differ": [0, 4, 8, 11, 18, 21, 23, 24, 35, 39, 42], "regularized_min_max_diff": [0, 12, 14], "regular": [0, 3, 13, 15], "gaussian": [0, 12], "divers": [0, 14], "averag": [0, 3, 4, 5, 12, 13, 14, 15], "same": [0, 35, 39], "per": [0, 2, 18, 21, 24, 31, 39, 40, 42], "first_layer_multipli": 0, "first": [0, 18, 21, 24, 26, 28, 31, 35, 42], "all": [0, 2, 3, 5, 8, 37, 40, 42], "other": [0, 13, 15], "grad": 0, "gradient": [0, 11, 28, 43], "base": [0, 3, 4, 9, 11, 13, 15, 17, 22, 25, 28, 29, 39, 40, 43], "manipul": 0, "random_crop": [0, 14], "crop": 0, "random_crop_flip": [0, 12], "flip": 0, "ident": 0, "do": [0, 42], "transform": [0, 18, 21, 24, 26, 28, 31], "torchvis": [0, 14, 26, 27, 28, 29, 30, 31, 35], "keras_appl": [0, 12], "imagenet": 0, "no_norm": 0, "known_dict": 1, "default_valu": 1, "dictionari": [1, 2, 3, 23, 24, 35, 37, 38, 40], "It": [1, 39, 40], "wrap": [1, 2, 8, 20, 24, 31, 36, 39, 40], "given": [1, 10, 18, 19, 21, 24, 26, 27, 28, 31], "return": [1, 3, 4, 7, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 35], "its": [1, 2, 8, 20, 22, 29, 36, 39, 42], "request": 1, "If": [1, 2, 4, 18, 21, 23, 24, 26, 28, 33, 35, 36, 39], "kei": [1, 5, 22, 29, 36], "present": [1, 42], "pass": [1, 2, 8, 13, 15, 18, 21, 22, 23, 24, 26, 28, 29, 30, 31, 37], "gener": [1, 11, 12, 14, 18, 19, 21, 22, 23, 24, 26, 27, 28, 29, 30, 31, 39, 42, 43], "provid": [1, 4, 17, 22, 25, 29, 35, 39, 40, 42], "empti": 1, "get": [1, 2, 3, 4, 11, 18, 21, 23, 24, 26, 28, 30, 31, 39, 42], "inner": 1, "default_factori": 1, "exist": [1, 37], "wa": [1, 35], "follow": [2, 3, 40, 42], "api": [2, 3, 21, 24, 31], "can": [2, 3, 5, 8, 11, 13, 15, 16, 17, 19, 22, 25, 27, 29, 34, 35, 37, 39, 40, 42, 43], "mct": [2, 8, 11, 13, 15, 16, 17, 18, 19, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 33, 34, 35, 37, 39, 40, 41, 42, 43], "framework": [2, 16, 39, 40], "relat": [2, 7, 11, 39], "inform": [2, 3, 11, 13, 15, 16, 18, 21, 22, 24, 26, 28, 29, 31, 34, 39, 40, 41], "network": [2, 6, 8, 30, 33, 34, 37, 42, 43], "core": [2, 4, 5, 8, 9, 10, 18, 19, 21, 22, 23, 24, 26, 27, 29, 30, 31, 33, 34, 37], "activation_quantizer_map": 2, "kernel_channels_map": 2, "activation_min_max_map": 2, "layer_min_max_map": 2, "kernel_ops_attributes_map": 2, "out_channel_axis_map": 2, "A": [2, 3, 4, 7, 11, 13, 15, 16, 18, 19, 20, 21, 22, 23, 24, 26, 27, 28, 29, 30, 31, 33, 34, 37, 38, 39, 43], "about": [2, 3, 7, 11, 13, 15, 18, 21, 23, 24, 35, 39, 40], "specif": [2, 22, 29, 37, 42], "librari": [2, 8], "need": [2, 18, 21, 24, 26, 28, 31, 35, 36, 40], "quantiz": [2, 3, 4, 8, 9, 10, 11, 13, 15, 17, 19, 25, 27, 33, 34, 37, 38, 39, 40, 42, 43], "model": [2, 3, 4, 7, 8, 10, 11, 16, 17, 18, 21, 22, 25, 26, 28, 29, 33, 37, 38, 39, 40, 42], "hold": [2, 33, 36, 39], "how": [2, 6, 18, 19, 21, 24, 26, 28, 31, 35, 40, 43], "thei": 2, "should": [2, 3, 6, 8, 10, 13, 15, 18, 19, 21, 22, 23, 24, 26, 28, 29, 31, 35, 39, 42], "multipl": [2, 4, 39], "map": [2, 39], "kernel": [2, 10, 18, 21, 23, 24, 37, 40], "channel": [2, 6, 7, 8, 11, 22, 29, 39, 40, 42], "indic": [2, 7, 22, 29, 39], "etc": [2, 10, 11, 18, 21, 24, 26, 28, 31, 42], "ar": [2, 3, 16, 18, 21, 22, 24, 26, 28, 29, 31, 35, 39, 40, 41, 42], "divid": 2, "three": 2, "group": [2, 6, 22, 29, 39], "kernel_op": 2, "have": [2, 35, 36, 42], "coeffici": [2, 10, 18, 21, 23, 24, 26, 28, 39, 40], "e": [2, 18, 21, 24, 26, 28, 31, 43], "g": [2, 18, 21, 24, 26, 28, 31], "conv2d": [2, 10, 17, 18, 21, 23, 24, 25, 37, 39], "dens": [2, 17], "activation_op": 2, "relu": [2, 8], "no_quantization_op": 2, "reshap": [2, 17], "transpos": 2, "dict": [2, 3, 7, 35, 39, 40], "quantizationmethod": [2, 40], "callabl": [2, 3, 4, 13, 15, 18, 19, 21, 22, 24, 26, 27, 28, 29, 31, 35, 36], "from": [2, 3, 8, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 35, 37, 39, 40, 41, 42, 43], "function": [2, 3, 4, 5, 11, 12, 13, 14, 15, 17, 20, 22, 25, 29, 37, 39, 40], "defaultdict": [2, 11, 39], "tupl": [2, 17, 18, 22, 29, 37], "out": [2, 6], "str": [2, 35, 36, 39], "activ": [2, 8, 10, 18, 19, 21, 24, 26, 27, 28, 31, 35, 37, 38, 39, 40, 42], "oper": [2, 10, 36, 39], "attirbut": 2, "s": [2, 6, 8, 10, 18, 21, 22, 23, 24, 26, 28, 29, 31, 35, 36, 37, 39, 40, 43], "comput": [2, 3, 4, 5, 9, 11, 19, 27, 42], "statist": [2, 18, 21, 24, 26, 28, 31, 42], "exampl": [2, 3, 8, 10, 13, 15, 17, 18, 19, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 37, 39, 40, 43], "kera": [2, 3, 10, 11, 37, 40, 43], "we": [2, 17, 18, 21, 22, 24, 25, 29, 31, 35, 37, 39, 40, 42], "want": 2, "onli": [2, 5, 6, 10, 18, 21, 23, 24, 35], "set": [2, 5, 11, 13, 17, 18, 21, 22, 23, 24, 25, 26, 28, 29, 31, 32, 37, 39, 40, 42], "know": 2, "3": [2, 13, 15, 17, 18, 19, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 35, 40], "2": [2, 8, 13, 17, 25, 39, 40, 43], "respectivli": 2, "import": [2, 6, 7, 8, 11, 13, 15, 17, 18, 19, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 35, 37, 40, 42], "tensorflow": [2, 11, 13, 16, 17, 18, 19, 21, 22, 23, 24, 35, 37, 39, 43], "tf": [2, 13, 17, 20, 23, 24], "Then": [2, 18, 21, 24, 26, 28, 31, 37, 42], "creat": [2, 3, 8, 11, 12, 13, 14, 15, 17, 18, 19, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 35, 36, 37], "object": [2, 6, 10, 12, 13, 14, 15, 16, 18, 19, 20, 21, 23, 24, 26, 27, 28, 31, 35, 37, 39, 40], "rang": [2, 18, 21, 24, 26, 28, 31], "advanc": 2, "save": [2, 24, 32, 35, 40], "collect": [2, 18, 21, 24, 26, 28, 31, 42], "time": [2, 6, 40], "For": [2, 8, 16, 17, 18, 21, 23, 24, 25, 31, 35, 39, 40, 41, 42, 43], "softmax": [2, 8], "1": [2, 3, 4, 7, 15, 17, 18, 19, 21, 22, 23, 25, 26, 27, 28, 29, 30, 35, 43], "select": [2, 6, 8, 9, 11, 35, 38, 39, 40], "format": [2, 11], "index": [2, 11], "axi": [2, 40], "nhwc": 2, "last": [2, 5], "nchw": 2, "instanc": [3, 11, 13, 15, 37, 39, 42], "which": [3, 6, 8, 35, 36, 37, 39, 40], "post": [3, 11, 22, 24, 29, 31, 43], "train": [3, 11, 38, 40, 43], "knowledg": [3, 43], "distil": [3, 43], "teacher": 3, "student": 3, "gptq": [3, 13, 15, 18, 26], "n_epoch": [3, 13, 15, 18], "optimizer_rest": [3, 13, 15], "log_funct": [3, 13, 15], "train_bia": 3, "rounding_typ": 3, "softquant": 3, "use_hessian_based_weight": [3, 13, 15], "optimizer_quantization_paramet": 3, "optimizer_bia": 3, "regularization_factor": [3, 13, 15], "reg_default": [3, 13, 15], "hessian_weights_config": 3, "gptq_quantizer_params_overrid": 3, "configur": [3, 4, 8, 10, 11, 12, 13, 14, 15, 17, 18, 21, 22, 23, 24, 25, 26, 28, 29, 30, 31, 33, 34, 36, 37, 38, 39, 40, 43], "gradientptq": [3, 11], "repres": [3, 4, 10, 13, 15, 18, 21, 22, 23, 24, 26, 28, 29, 30, 31, 35, 37, 39, 42], "dataset": [3, 13, 15, 18, 19, 21, 22, 23, 24, 26, 27, 28, 29, 30, 31, 35, 42], "epoch": [3, 13, 15], "bia": [3, 8, 10, 13, 15, 18, 21, 23, 24], "accept": [3, 13, 15], "6": [3, 25], "tensor": [3, 4, 13, 15, 17, 19, 25, 27, 39, 40, 42, 43], "1st": [3, 13, 15], "2nd": [3, 13, 15], "3rd": [3, 13, 15], "4th": [3, 13, 15], "5th": 3, "6th": 3, "mean": [3, 9, 42], "std": 3, "accordingli": [3, 39], "see": [3, 43], "multiple_tensors_mse_loss": [3, 15], "log": [3, 11, 13, 15, 32, 42], "process": [3, 4, 8, 11, 12, 13, 14, 15, 16, 17, 22, 25, 29, 33, 34, 37, 38, 39, 41, 42], "whether": [3, 4, 7, 8, 12, 13, 14, 15, 20, 34, 35, 39, 40], "updat": 3, "dure": [3, 8, 11, 12, 13, 14, 15, 16, 35, 37, 39, 40, 41, 42], "defin": [3, 4, 5, 13, 15, 17, 18, 21, 22, 23, 24, 25, 26, 28, 29, 39, 40], "round": 3, "hessian": [3, 4, 6, 9, 13, 15, 22, 29, 43], "overrid": [3, 38], "rest": 3, "point": [3, 4, 13, 15, 18, 26, 28, 39, 42], "factor": [3, 4, 9, 13, 15], "includ": [3, 7, 10, 18, 21, 24, 26, 28, 31, 38], "necessari": [3, 35, 40], "argument": [3, 5, 39], "run": [3, 13, 15, 35, 42], "score": [3, 4, 6, 7, 8, 9, 22, 29], "instanti": [3, 8, 38], "hessians_num_sampl": 3, "16": [3, 35], "norm_scor": [3, 4], "log_norm": 3, "scale_log_norm": 3, "fals": [3, 4, 8, 34, 39], "metric": [3, 4, 5, 6, 10, 11], "gptqhessianweightsconfig": 3, "sampl": [3, 42], "between": [3, 4, 8, 18, 26, 28, 39, 42], "scale": [3, 4, 8, 39], "final": [3, 4, 11, 17, 25, 37, 42, 43], "vector": [3, 42], "method": [3, 4, 5, 6, 8, 9, 11, 22, 29, 35, 37, 38, 39, 40], "ste": [3, 38], "straight": [3, 38], "through": [3, 17, 22, 25, 38], "estim": [3, 38], "mix": [4, 5, 10, 11, 18, 19, 21, 23, 24, 26, 27, 28, 31, 33, 39, 43], "precis": [4, 5, 10, 11, 18, 19, 21, 22, 23, 24, 26, 27, 28, 29, 31, 33, 39, 43], "compute_distance_fn": 4, "distance_weighting_method": 4, "mpdistanceweight": [4, 11], "avg": [4, 5], "num_of_imag": [4, 18, 21], "32": 4, "configuration_overwrit": 4, "num_interest_points_factor": 4, "use_hessian_based_scor": 4, "refine_mp_solut": 4, "metric_normalization_threshold": 4, "1e10": 4, "distanc": [4, 5, 11], "two": [4, 18, 21, 24, 26, 28, 31, 35, 39, 42], "pre": 4, "each": [4, 6, 7, 18, 21, 22, 24, 26, 28, 29, 31, 37, 39, 40, 42], "among": 4, "sensit": [4, 6, 22, 29], "evalu": 4, "compar": [4, 18, 26, 28, 42], "integ": [4, 35, 39], "overwrit": 4, "predefin": [4, 6], "one": [4, 8, 36, 42], "zero": [4, 39], "percentag": 4, "reduc": [4, 22, 29], "interest": 4, "calcul": [4, 6, 11, 18, 19, 21, 22, 24, 26, 27, 28, 29, 31], "try": 4, "improv": [4, 22, 29], "greedi": [4, 6], "algorithm": 4, "search": [4, 8, 10, 11, 18, 21, 24, 26, 28, 31], "increas": 4, "bit": [4, 10, 18, 21, 23, 24, 31, 35, 37, 39, 40, 43], "width": [4, 18, 21, 24, 25, 31, 39, 43], "threshold": [4, 8, 9, 18, 21, 24, 26, 28, 31, 39, 40], "check": [4, 35, 36, 37], "In": [4, 17, 18, 21, 24, 25, 26, 28, 31, 35, 36], "case": 4, "larger": 4, "than": [4, 36], "thi": [4, 7, 9, 10, 11, 17, 18, 20, 21, 22, 23, 24, 25, 26, 28, 29, 31, 35, 39, 40, 43], "prevent": 4, "numer": 4, "issu": [4, 35], "call": [5, 19, 27, 39, 42], "take": [5, 21, 24, 31, 43], "last_lay": 5, "pruningconfig": [6, 11, 22, 29], "num_score_approxim": [6, 22, 29], "pruning_num_score_approxim": 6, "importance_metr": 6, "lfh": [6, 22, 29], "channels_filtering_strategi": 6, "specifi": [6, 12, 14, 17, 20, 22, 25, 29, 35], "neural": [6, 43], "approxim": [6, 22, 29], "perform": [6, 10, 17, 22, 25, 29], "strategi": [6, 22, 29], "filter": [6, 39], "constant": [6, 37, 40], "label": [6, 22, 29, 39, 43], "free": [6, 17, 22, 25, 29, 43], "approach": 6, "determin": [6, 22, 29], "info": [6, 32], "measur": [6, 10, 42], "least": 6, "up": [6, 17, 25, 39, 42], "allow": [6, 17, 25, 35], "resourc": [6, 10, 11, 18, 21, 22, 23, 24, 29, 30, 31, 42], "util": [6, 10, 11, 18, 21, 22, 23, 24, 29, 30, 31, 40], "limit": [6, 18, 21, 23, 24, 26, 28, 31], "now": [6, 16, 31, 35, 39, 40, 41, 42], "weights_memori": [6, 10, 18, 21, 22, 24, 29, 31], "consid": [6, 12, 14, 22, 29, 39], "pruninginfo": [7, 11, 22, 29], "pruning_mask": 7, "importance_scor": 7, "store": [7, 40], "mask": 7, "act": 7, "contain": [7, 11, 17, 18, 19, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 39, 40], "access": 7, "metadata": [7, 39], "basenod": 7, "np": [7, 10, 18, 19, 21, 22, 23, 24, 26, 27, 28, 29, 30, 31, 35], "ndarrai": 7, "arrai": 7, "where": [7, 35, 37, 39, 42], "element": [7, 39], "correspond": 7, "neuron": 7, "ha": [7, 35, 36, 37], "been": 7, "kept": [7, 24, 31], "quantifi": [7, 42], "signific": 7, "properti": 7, "activation_error_method": 8, "quantizationerrormethod": [8, 11], "mse": [8, 9, 42], "weights_error_method": 8, "relu_bound_to_power_of_2": 8, "weights_bias_correct": 8, "weights_second_moment_correct": 8, "input_sc": 8, "softmax_shift": 8, "shift_negative_activation_correct": 8, "activation_channel_equ": 8, "z_threshold": 8, "math": 8, "inf": [8, 10], "min_threshold": [8, 40], "l_p_valu": 8, "linear_collaps": 8, "residual_collaps": 8, "shift_negative_ratio": 8, "05": 8, "shift_negative_threshold_recalcul": 8, "shift_negative_params_search": 8, "concat_threshold_upd": 8, "accord": [8, 10, 11, 18, 19, 21, 22, 24, 26, 27, 28, 29, 31, 35, 36, 39], "power": [8, 18, 21, 24, 26, 28, 31, 39], "correct": 8, "second_mo": 8, "shift": 8, "neg": 8, "equal": [8, 36], "z": 8, "outlier": 8, "remov": [8, 22, 29, 30], "minimum": [8, 40], "p": [8, 29], "l_p": 8, "norm": [8, 9, 42], "block_collaps": 8, "collaps": 8, "block": [8, 40, 42], "anoth": 8, "ratio": 8, "minim": [8, 9, 18, 22, 26, 28, 29], "non": [8, 39], "linear": [8, 25], "abov": 8, "occur": 8, "recomput": 8, "after": [8, 11, 18, 20, 21, 24, 31, 43], "One": [8, 42], "mai": [8, 17, 18, 21, 24, 25, 26, 28, 31, 36, 42], "done": [8, 42], "noclip": [8, 9], "qc": 8, "instans": 8, "keras_post_training_quant": [8, 11, 13, 21, 35, 37, 42], "error": 9, "nois": 9, "mae": [9, 42], "absolut": 9, "kl": [9, 42], "diverg": [9, 42], "make": 9, "signal": 9, "distribut": 9, "similar": [9, 34, 43], "possibl": [9, 18, 21, 24, 31, 39, 42], "lp": 9, "hmse": 9, "more": [9, 16, 21, 22, 24, 29, 31, 35, 39, 41, 42], "valuabl": 9, "induc": 9, "prune": [10, 43], "activation_memori": 10, "total_memori": 10, "bop": 10, "memori": [10, 22, 29, 42], "byte": [10, 18, 21, 22, 24, 29, 31, 42], "note": [10, 18, 21, 23, 24], "affect": [10, 18, 21, 23, 24], "while": [10, 18, 21, 23, 24, 31, 39], "sum": [10, 19, 22, 27, 29], "total": [10, 19, 27], "init": [11, 37, 43], "modul": [11, 25, 26, 27, 28, 29], "pytorch_post_training_quant": [11, 15, 28, 35], "pytorch": [11, 16, 39, 40, 43], "pytorch_gradient_post_training_quant": [11, 26], "get_pytorch_gptq_config": [11, 15], "gradientptqconfig": [11, 18, 26], "keras_gradient_post_training_quant": [11, 18], "get_keras_gptq_config": [11, 13, 18], "option": [11, 20, 21, 22, 24, 26, 29, 31, 35, 39], "pytorch_quantization_aware_training_init_experiment": [11, 30, 31], "prepar": [11, 24, 31], "awar": [11, 38, 40, 43], "experiment": [11, 17, 25, 43], "pytorch_quantization_aware_training_finalize_experiment": [11, 30], "without": 11, "quantizewrapp": [11, 24, 30, 31], "keras_quantization_aware_training_init_experiment": [11, 23, 24], "keras_quantization_aware_training_finalize_experiment": [11, 23], "qat_config": [11, 24, 31], "coreconfig": [11, 18, 19, 21, 23, 24, 26, 27, 28, 30, 31], "entir": 11, "quantizationconfig": [11, 33], "mixedprecisionquantizationconfig": [11, 18, 19, 21, 23, 24, 33], "resourceutil": [11, 18, 19, 21, 22, 23, 24, 26, 27, 28, 29, 31], "network_editor": [11, 34], "modifi": [11, 37], "troubleshoot": 11, "pytorch_resource_utilization_data": [11, 27], "data": [11, 12, 14, 19, 22, 27, 29, 35, 39, 42, 43], "desir": [11, 18, 19, 21, 23, 24, 26, 27, 28, 31], "target": [11, 16, 18, 19, 21, 22, 23, 24, 27, 29, 30, 31], "keras_resource_utilization_data": [11, 19], "pytorch_data_generation_experiment": [11, 25], "get_pytorch_data_generation_config": [11, 14, 25], "load": [11, 23, 24, 35, 40], "keras_data_generation_experiment": [11, 17], "get_keras_data_generation_config": [11, 12, 17], "pytorch_pruning_experiment": [11, 29], "structur": [11, 43], "keras_pruning_experiment": [11, 22], "serial": 11, "abstract": [11, 40], "hardwar": [11, 16, 22, 29, 39, 40, 43], "orient": [11, 40], "tool": [11, 40], "logger": [11, 34, 42], "path": [11, 20, 32, 35, 42], "directori": [11, 32], "infer": [11, 16, 23, 30, 39, 40], "get_target_platform_cap": [11, 16, 39], "platform": [11, 16, 18, 21, 22, 23, 24, 27, 29], "targetplatformcap": [11, 18, 19, 21, 22, 24, 26, 27, 28, 29, 31], "page": 11, "document": [11, 21, 24, 31], "auto": 11, "sphinx": 11, "default_n_it": [12, 14], "adam": [12, 13, 15], "default_data_gen_b": [12, 14], "default_keras_initial_lr": 12, "default_keras_output_loss_multipli": 12, "schedulertyp": [12, 14], "batchnormalignemntlosstyp": [12, 14], "outputlosstyp": [12, 14], "datainittyp": [12, 14], "bnlayerweightingtyp": [12, 14], "imagegranular": [12, 14], "imagepipelinetyp": [12, 14], "imagenormalizationtyp": [12, 14], "learning_r": 13, "lr_default": [13, 15], "lr_rest_default": [13, 15], "gptqmultipletensorsloss": 13, "gradientptqconfigv2": [13, 15], "fine": [13, 15, 22, 23, 24, 29, 30, 31], "tune": [13, 15, 22, 23, 24, 29, 30], "optimizerv2": 13, "auxiliri": [13, 15], "variabl": [13, 15], "4": [13, 15, 17, 18, 21, 22, 24, 25, 26, 28, 29, 31], "5": [13, 15, 22, 29], "gptq_conf": [13, 15, 26], "nadam": 13, "order": [13, 15, 18, 21, 24, 31, 35, 36], "radam": 14, "default_pytorch_initial_lr": 14, "default_pytorch_output_loss_multipli": 14, "default_pytorch_bn_layer_typ": 14, "default_pytorch_last_layer_typ": 14, "torch": [15, 25, 35, 43], "lr": 15, "dummi": 15, "param": [15, 34, 37, 40], "fw_name": 16, "target_platform_nam": 16, "target_platform_vers": 16, "name": [16, 35, 37, 39, 42], "support": [16, 35], "both": [16, 18, 21, 26, 28, 30, 40, 42], "them": [16, 39, 42], "imx500": [16, 35, 39], "tflite": [16, 35, 39], "qnnpack": [16, 39], "capabl": [16, 22, 27, 29], "version": [16, 17, 25, 39, 43], "attach": [16, 39], "some": [16, 17, 25, 35, 39, 41, 42], "field": [16, 36, 39, 41], "opquantizationconfig": [16, 41], "ignor": [16, 39, 41], "quantization_preserv": [16, 39, 41], "fixed_scal": [16, 39, 41], "fixed_zero_point": [16, 39, 41], "futur": [16, 17, 25, 39, 41], "n_imag": [17, 25], "output_image_s": [17, 25], "data_generation_config": [17, 25], "ll": [17, 25], "walk": [17, 25], "simpl": [17, 25], "involv": [17, 22, 25, 29], "start": [17, 25, 35, 40, 43], "compress": [17, 22, 25, 26, 29], "toolkit": [17, 25, 26], "sequenti": [17, 25], "flatten": [17, 25], "next": [17, 25, 35, 36], "input_shap": 17, "8": [17, 18, 21, 23, 24, 25, 35, 40], "10": [17, 18, 21, 24, 25, 26, 28, 31], "custom": [17, 20, 24, 25, 35], "simplic": [17, 25], "config": [17, 18, 21, 22, 23, 24, 25, 26, 29, 30, 31, 33, 37, 39, 40], "notic": [17, 22, 25, 29, 35], "chang": [17, 25, 35, 37, 42], "generated_imag": [17, 25], "variou": [17, 25, 42], "purpos": [17, 25, 34], "in_model": [18, 19, 21, 23, 24, 27, 30, 31], "representative_data_gen": [18, 19, 21, 22, 24, 26, 27, 28, 29, 31, 35], "gptq_config": [18, 26, 28], "gptq_representative_data_gen": [18, 26], "target_resource_util": [18, 21, 22, 24, 26, 28, 29, 31], "core_config": [18, 19, 21, 23, 24, 26, 27, 28, 30, 31], "target_platform_cap": [18, 19, 21, 22, 24, 26, 27, 28, 29, 31, 40], "default_keras_tpc": [18, 21, 22, 24], "symmetr": [18, 21, 24, 26, 28, 31, 38, 39, 40], "constraint": [18, 21, 22, 26, 28, 29], "sever": [18, 21, 24, 26, 28, 31, 42], "fold": [18, 21, 24, 26, 28, 31], "preced": [18, 21, 24, 26, 28, 31], "histogram": [18, 21, 24, 26, 28, 31, 42], "being": [18, 21, 24, 26, 28, 31, 39, 40], "ilp": [18, 21, 24, 31], "solver": [18, 21, 24, 31], "find": [18, 21, 24, 31], "maxim": [18, 21, 24, 31], "observ": [18, 26, 28, 39, 42], "calibr": [18, 19, 21, 24, 26, 27, 28, 31], "user": [18, 21, 23, 24, 26, 28, 30, 31], "handl": [18, 21, 24, 26, 28, 31], "applic": [18, 19, 21, 22, 23, 24, 35], "mobilenet": [18, 19], "random": [18, 19, 21, 22, 23, 24, 26, 27, 28, 29, 30, 31, 35], "requir": [18, 21, 24, 26, 28, 31, 40, 42], "num_calibration_batch": [18, 21, 24, 26, 28, 31], "numpi": [18, 19, 21, 22, 23, 24, 26, 27, 28, 29, 30, 31, 35], "def": [18, 19, 21, 22, 23, 24, 26, 27, 28, 29, 30, 31, 35], "repr_datagen": [18, 19, 21, 22, 23, 24, 26, 27, 28, 29, 30, 31], "_": [18, 21, 24, 26, 28, 31, 35], "yield": [18, 19, 21, 22, 23, 24, 26, 27, 28, 29, 30, 31, 35], "224": [18, 19, 21, 22, 23, 24, 26, 27, 28, 29, 30, 31, 35], "bitwidth": [18, 21, 23, 24], "candid": [18, 21, 23, 24, 37], "mixed_precision_config": [18, 19, 21, 23, 24, 33], "our": [18, 21, 23, 24, 31, 43], "ru": [18, 21, 23, 24], "count_param": [18, 21, 22, 23, 24], "75": [18, 21, 23, 24], "quantized_model": [18, 21, 23, 24, 30, 31], "quantization_info": [18, 21, 23, 24, 26, 28, 30, 31], "userinform": 18, "keras_default_tpc": 19, "build": [19, 27, 40, 43], "graph": [19, 27, 37, 42], "hw": 19, "ru_data": [19, 27], "keras_load_quantized_model": 20, "filepath": 20, "custom_object": [20, 23, 24], "compil": 20, "trainabl": [20, 23, 40], "file": [20, 23, 24, 32, 35], "addit": [20, 35], "boolean": 20, "saved_model": 20, "loadopt": 20, "savedmodel": 20, "ptq": [21, 28, 35], "mobilenet_v2": [21, 23, 24, 26, 27, 28, 30, 31, 35], "mobilenetv2": [21, 23, 24, 35, 42], "pleas": [21, 24, 31, 35, 43], "look": [21, 24, 31, 39, 43], "pruning_config": [22, 29], "meet": [22, 29], "simd": [22, 29, 39], "tpc": [22, 29, 39], "By": [22, 26, 28, 29, 35, 42], "assess": [22, 29], "togeth": [22, 29], "friendli": [22, 29, 43], "architectur": [22, 29], "analyz": [22, 29], "identifi": [22, 29], "impact": [22, 29], "must": [22, 29, 39], "retrain": [22, 29], "recov": [22, 29], "origin": [22, 42], "achiev": 22, "analysi": [22, 29], "standard": [22, 29, 38], "associ": [22, 29, 39], "resnet50": [22, 29, 35], "here": [22, 29, 35, 39, 43], "aim": [22, 29], "footprint": [22, 29], "50": [22, 29], "assum": [22, 29], "float32": [22, 29, 35], "thu": [22, 29, 42], "dense_nparam": [22, 29], "l": [22, 43], "higher": [22, 29], "also": [22, 29, 43], "extend": [22, 29], "durat": [22, 29], "pruned_model": [22, 29], "pruning_info": [22, 29], "qat": [23, 24, 30, 31, 38], "convert": [23, 30], "replac": 23, "trainablequant": 23, "inferablequant": [23, 30], "load_model": [23, 24], "model_fil": [23, 24], "qatconfig": [24, 31], "built": [24, 31, 40], "fake_qu": [24, 31], "node": [24, 31, 37, 40, 42], "onlin": [24, 31], "wrapper": [24, 30, 31, 40], "ker": 24, "hight": 25, "nn": 25, "batchnorm2d": 25, "default_pytorch_tpc": [26, 28, 29, 31], "quantized_modul": [26, 28], "pytorch_default_tpc": 27, "in_modul": 28, "clibrat": 28, "default_pyotrch_tpc": 29, "resnet50_weight": 29, "imagenet1k_v1": 29, "numel": 29, "state_dict": 29, "pretrain": [30, 31], "keep": [30, 43], "readi": 30, "tun": 31, "set_log_fold": [32, 42], "folder": 32, "level": 32, "verbos": 32, "quantization_config": [33, 40], "debug_config": 33, "debugconfig": 33, "debug": [33, 34], "edit": [33, 34, 37], "analyze_similar": 34, "plot": [34, 42], "figur": [34, 42], "within": [34, 43], "tensorboard": [34, 43], "pinpoint": 34, "problemat": 34, "editrul": 34, "rule": [34, 37], "action": 34, "detail": [35, 39], "refer": 35, "project": [35, 43], "github": [35, 43], "readm": 35, "you": [35, 42, 43], "question": 35, "open": [35, 42, 43], "repositori": 35, "fakely_qu": 35, "int8": 35, "mct_quantiz": 35, "via": [35, 43], "save_model_path": 35, "is_layer_exportable_fn": 35, "is_keras_layer_export": 35, "serialization_format": 35, "quantization_format": 35, "To": [35, 42], "float_model": 35, "demonstr": [35, 39], "quantized_exportable_model": 35, "lambda": 35, "dtype": 35, "avail": [35, 43], "tempfil": 35, "keras_file_path": 35, "mkstemp": 35, "torchscript": 35, "repr_dataset": 35, "is_pytorch_layer_export": 35, "onnx_opset_vers": 35, "default_onnx_opset_vers": 35, "current": 35, "fake": 35, "trace": 35, "mandatori": 35, "quant": 35, "your": 35, "packag": [35, 40, 43], "so": [35, 39], "part": 35, "skip": 35, "plan": 35, "pip": [35, 43], "instal": 35, "q": 35, "onnxruntim": 35, "extens": 35, "let": 35, "There": [35, 42], "onnx_file_path": 35, "model_format_onnx_mctq": 35, "15": 35, "get_ort_session_opt": 35, "session": 35, "creation": 35, "slowli": 35, "suffer": 35, "longer": 35, "latenc": 35, "howev": 35, "ort": 35, "sess": 35, "inferencesess": 35, "cudaexecutionprovid": 35, "cpuexecutionprovid": 35, "_input_data": 35, "astyp": 35, "_model_output_nam": 35, "get_output": 35, "_model_input_nam": 35, "get_input": 35, "predict": 35, "layerfilterparam": 36, "target_platform": [36, 40], "attributefilt": [36, 39], "attr": 36, "op": [36, 39], "match": [36, 37, 39], "regard": 36, "eq": 36, "noteq": 36, "greater": 36, "greatereq": 36, "smaller": 36, "smallereq": 36, "attribut": [37, 39, 40], "9": 37, "er_list": 37, "nodetypefilt": 37, "changecandidatesweightsquantconfigattr": 37, "attr_nam": 37, "weights_n_bit": [37, 39, 40], "new": 37, "node_typ": 37, "nodenamefilt": 37, "node_nam": 37, "nodenamescopefilt": 37, "node_name_scop": 37, "string": 37, "changefinalweightsquantconfigattr": 37, "kwarg": [37, 39], "attr_valu": 37, "changefinalactivationquantconfigattr": 37, "changecandidatesactivationquantconfigattr": 37, "changequantizationparamfunct": 37, "activation_quantization_params_fn": 37, "weights_quantization_params_fn": 37, "changefinalweightsquantizationmethod": 37, "weights_quantization_method": [37, 39, 40], "changecandidatesweightsquantizationmethod": 37, "changecandidatesactivationquantizationmethod": 37, "activation_quantization_method": [37, 39, 40], "changequantizationmethod": 37, "poweroftwo": 38, "uniform": [38, 39, 40], "dqa": 38, "dnn": 38, "attent": 38, "smooth": 38, "introduc": 38, "lsq": 38, "http": [38, 43], "arxiv": [38, 43], "org": 38, "pdf": 38, "1902": 38, "08153": 38, "weight_training_method": 38, "activation_training_method": 38, "weight_quantizer_params_overrid": 38, "activation_quantizer_params_overrid": 38, "backend": 39, "address": 39, "shortli": 39, "diagram": 39, "main": [39, 42], "compon": [39, 40], "power_of_two": 39, "lut_pot_quant": 39, "lookup": 39, "tabl": 39, "lut_sym_quant": 39, "default_weight_attr_config": 39, "attr_weights_configs_map": 39, "activation_n_bit": [39, 40], "enable_activation_quant": [39, 40], "simd_siz": 39, "singl": 39, "instruct": 39, "fetch": 39, "simultan": 39, "float_bitwidth": 39, "weights_per_channel_threshold": [39, 40], "enable_weights_quant": [39, 40], "lut_values_bitwidth": 39, "quantization_config_list": 39, "base_config": 39, "gather": [39, 42], "fallback": 39, "manner": 39, "default_qco": 39, "add_metadata": 39, "default_tp_model": 39, "definit": 39, "pattern": 39, "combin": 39, "qc_option": 39, "uniqu": 39, "operator_groups_list": 39, "treat": 39, "henc": 39, "union": 39, "either": 39, "opset": 39, "concaten": 39, "similarli": 39, "place": 39, "like": 39, "op_sets_to_lay": 39, "op_set_nam": 39, "attr_map": 39, "filterlayerparam": 39, "condit": 39, "satisfi": 39, "keyword": 39, "usag": 39, "tp_model": 39, "infrastructur": 40, "propos": 40, "emul": 40, "holder": 40, "upon": 40, "learnabl": 40, "explain": 40, "basic": 40, "valid": 40, "implement": 40, "get_config": 40, "from_config": 40, "weights_quantization_param": 40, "weights_channels_axi": 40, "weights_quantization_candid": 40, "activation_quantization_param": 40, "activation_quantization_candid": 40, "phase": 42, "displai": 42, "ui": 42, "dir": 42, "writer": 42, "stage": 42, "launch": 42, "logdir": 42, "seen": 42, "under": 42, "tab": 42, "click": 42, "deeper": 42, "view": 42, "tag": 42, "show": 42, "These": 42, "scalar": 42, "were": 42, "wai": [42, 43], "mathemat": 42, "divis": 42, "dot": 42, "product": 42, "euclidean": 42, "along": 42, "pair": 42, "would": 42, "expect": 42, "mani": 42, "few": [42, 43], "As": 42, "20": 42, "insert": 42, "chosen": 42, "second": 42, "sourc": 43, "research": 43, "develop": 43, "engin": 43, "easili": 43, "state": 43, "art": 43, "work": 43, "soni": 43, "semiconductor": 43, "israel": 43, "git": 43, "clone": 43, "com": 43, "model_optim": 43, "python": 43, "setup": 43, "py": 43, "pypi": 43, "latest": 43, "stabl": 43, "releas": 43, "nightli": 43, "unstabl": 43, "visual": 43, "cosin": 43, "comparison": 43, "just": 43, "minut": 43, "visit": 43, "notebook": 43, "doesn": 43, "t": 43, "out1": 43, "out2": 43, "out3": 43, "habi": 43, "h": 43, "v": 43, "peretz": 43, "r": 43, "cohen": 43, "dikstein": 43, "dror": 43, "o": 43, "diamant": 43, "i": 43, "jen": 43, "netzer": 43, "2021": 43, "hptq": 43, "preprint": 43, "gordon": 43, "2023": 43, "eptq": 43, "enhanc": 43}, "objects": {"model_compression_toolkit": [[1, 0, 1, "", "DefaultDict"], [16, 2, 1, "", "get_target_platform_capabilities"], [20, 2, 1, "", "keras_load_quantized_model"], [32, 2, 1, "", "set_log_folder"]], "model_compression_toolkit.DefaultDict": [[1, 1, 1, "", "get"], [1, 1, 1, "", "keys"]], "model_compression_toolkit.core": [[2, 0, 1, "", "ChannelAxis"], [33, 0, 1, "", "CoreConfig"], [34, 0, 1, "", "DebugConfig"], [2, 0, 1, "", "FrameworkInfo"], [4, 0, 1, "", "MixedPrecisionQuantizationConfig"], [5, 0, 1, "", "MpDistanceWeighting"], [8, 0, 1, "", "QuantizationConfig"], [9, 0, 1, "", "QuantizationErrorMethod"], [10, 0, 1, "", "ResourceUtilization"], [19, 2, 1, "", "keras_resource_utilization_data"], [27, 2, 1, "", "pytorch_resource_utilization_data"]], "model_compression_toolkit.core.network_editor": [[37, 0, 1, "", "ChangeCandidatesActivationQuantConfigAttr"], [37, 0, 1, "", "ChangeCandidatesActivationQuantizationMethod"], [37, 0, 1, "", "ChangeCandidatesWeightsQuantConfigAttr"], [37, 0, 1, "", "ChangeCandidatesWeightsQuantizationMethod"], [37, 0, 1, "", "ChangeFinalActivationQuantConfigAttr"], [37, 0, 1, "", "ChangeFinalWeightsQuantConfigAttr"], [37, 0, 1, "", "ChangeFinalWeightsQuantizationMethod"], [37, 0, 1, "", "ChangeQuantizationParamFunction"], [37, 0, 1, "", "EditRule"], [37, 0, 1, "", "NodeNameFilter"], [37, 0, 1, "", "NodeNameScopeFilter"], [37, 0, 1, "", "NodeTypeFilter"]], "model_compression_toolkit.data_generation": [[0, 0, 1, "", "BNLayerWeightingType"], [0, 0, 1, "", "BatchNormAlignemntLossType"], [0, 0, 1, "", "DataGenerationConfig"], [0, 0, 1, "", "DataInitType"], [0, 0, 1, "", "ImageGranularity"], [0, 0, 1, "", "ImageNormalizationType"], [0, 0, 1, "", "ImagePipelineType"], [0, 0, 1, "", "OutputLossType"], [0, 0, 1, "", "SchedulerType"], [12, 2, 1, "", "get_keras_data_generation_config"], [14, 2, 1, "", "get_pytorch_data_generation_config"], [17, 2, 1, "", "keras_data_generation_experimental"], [25, 2, 1, "", "pytorch_data_generation_experimental"]], "model_compression_toolkit.exporter": [[35, 0, 1, "", "KerasExportSerializationFormat"], [35, 0, 1, "", "PytorchExportSerializationFormat"], [35, 0, 1, "", "QuantizationFormat"], [35, 0, 1, "", "keras_export_model"], [35, 0, 1, "", "pytorch_export_model"]], "model_compression_toolkit.gptq": [[3, 0, 1, "", "GPTQHessianScoresConfig"], [3, 0, 1, "", "GradientPTQConfig"], [3, 0, 1, "", "RoundingType"], [13, 2, 1, "", "get_keras_gptq_config"], [15, 2, 1, "", "get_pytorch_gptq_config"], [18, 2, 1, "", "keras_gradient_post_training_quantization"], [26, 2, 1, "", "pytorch_gradient_post_training_quantization"]], "model_compression_toolkit.pruning": [[6, 0, 1, "", "ChannelsFilteringStrategy"], [6, 0, 1, "", "ImportanceMetric"], [6, 0, 1, "", "PruningConfig"], [7, 0, 1, "", "PruningInfo"], [22, 2, 1, "", "keras_pruning_experimental"], [29, 2, 1, "", "pytorch_pruning_experimental"]], "model_compression_toolkit.pruning.PruningConfig": [[6, 3, 1, "", "channels_filtering_strategy"], [6, 3, 1, "", "importance_metric"], [6, 3, 1, "", "num_score_approximations"]], "model_compression_toolkit.pruning.PruningInfo": [[7, 4, 1, "", "importance_scores"], [7, 4, 1, "", "pruning_masks"]], "model_compression_toolkit.ptq": [[21, 2, 1, "", "keras_post_training_quantization"], [28, 2, 1, "", "pytorch_post_training_quantization"]], "model_compression_toolkit.qat": [[38, 0, 1, "", "QATConfig"], [38, 0, 1, "", "TrainingMethod"], [23, 2, 1, "", "keras_quantization_aware_training_finalize_experimental"], [24, 2, 1, "", "keras_quantization_aware_training_init_experimental"], [30, 2, 1, "", "pytorch_quantization_aware_training_finalize_experimental"], [31, 2, 1, "", "pytorch_quantization_aware_training_init_experimental"]], "model_compression_toolkit.target_platform": [[36, 0, 1, "", "AttributeFilter"], [39, 0, 1, "", "AttributeQuantizationConfig"], [36, 0, 1, "", "Eq"], [39, 0, 1, "", "Fusing"], [36, 0, 1, "", "Greater"], [36, 0, 1, "", "GreaterEq"], [39, 0, 1, "", "LayerFilterParams"], [36, 0, 1, "", "NotEq"], [39, 0, 1, "", "OpQuantizationConfig"], [39, 0, 1, "", "OperationsSetToLayers"], [39, 0, 1, "", "OperationsToLayers"], [39, 0, 1, "", "OperatorSetConcat"], [39, 0, 1, "", "OperatorsSet"], [39, 0, 1, "", "QuantizationConfigOptions"], [39, 0, 1, "", "QuantizationMethod"], [36, 0, 1, "", "Smaller"], [36, 0, 1, "", "SmallerEq"], [39, 0, 1, "", "TargetPlatformCapabilities"], [39, 0, 1, "", "TargetPlatformModel"]], "model_compression_toolkit.trainable_infrastructure": [[40, 0, 1, "", "BaseKerasTrainableQuantizer"], [40, 0, 1, "", "BasePytorchTrainableQuantizer"], [40, 0, 1, "", "TrainableQuantizerActivationConfig"], [40, 0, 1, "", "TrainableQuantizerWeightsConfig"]]}, "objtypes": {"0": "py:class", "1": "py:method", "2": "py:function", "3": "py:attribute", "4": "py:property"}, "objnames": {"0": ["py", "class", "Python class"], "1": ["py", "method", "Python method"], "2": ["py", "function", "Python function"], "3": ["py", "attribute", "Python attribute"], "4": ["py", "property", "Python property"]}, "titleterms": {"data": [0, 17, 25], "gener": [0, 17, 25], "configur": [0, 6, 42], "imagegranular": 0, "schedulertyp": 0, "batchnormalignemntlosstyp": 0, "outputlosstyp": 0, "datainittyp": 0, "bnlayerweightingtyp": 0, "imagepipelinetyp": 0, "imagenormalizationtyp": 0, "defaultdict": 1, "class": [1, 2, 3], "frameworkinfo": 2, "channelaxi": 2, "gradientptqconfig": [3, 13, 15], "gptqhessianscoresconfig": 3, "roundingtyp": 3, "mixedprecisionquantizationconfig": 4, "mpdistanceweight": 5, "prune": [6, 7, 11, 22, 29], "importancemetr": 6, "channelsfilteringstrategi": 6, "inform": [7, 19, 27], "quantizationconfig": 8, "quantizationerrormethod": 9, "resourceutil": 10, "api": [11, 43], "doc": 11, "ptq": 11, "gptq": 11, "qat": 11, "core": 11, "data_gener": 11, "export": [11, 35], "trainable_infrastructur": [11, 40], "set_log_fold": 11, "keras_load_quantized_model": 11, "target_platform": [11, 39], "indic": 11, "tabl": 11, "get": [12, 13, 14, 15, 16, 19, 27], "datagenerationconfig": [12, 14], "kera": [12, 13, 17, 18, 19, 20, 21, 22, 23, 24, 35], "model": [12, 13, 14, 15, 19, 20, 23, 24, 27, 30, 31, 35, 43], "pytorch": [14, 15, 25, 26, 27, 28, 29, 30, 31, 35], "targetplatformcap": [16, 39], "gradient": [18, 26], "base": [18, 26], "post": [18, 21, 26, 28], "train": [18, 21, 23, 24, 26, 28, 30, 31], "quantiz": [18, 20, 21, 23, 24, 26, 28, 30, 31, 35], "resourc": [19, 27], "util": [19, 27], "load": 20, "structur": [22, 29], "awar": [23, 24, 30, 31], "final": [23, 30], "init": [24, 31], "enabl": 32, "logger": 32, "coreconfig": 33, "debug_config": 34, "modul": [34, 35, 37, 38, 39, 40], "debugconfig": 34, "quantizationformat": 35, "kerasexportserializationformat": 35, "keras_export_model": 35, "tutori": 35, "serial": 35, "format": 35, "mctq": 35, "pytorchexportserializationformat": 35, "pytorch_export_model": 35, "onnx": 35, "opset": 35, "version": 35, "us": 35, "infer": 35, "layer": 36, "attribut": 36, "filter": [36, 37], "network_editor": 37, "editrul": 37, "action": 37, "qat_config": 38, "trainingmethod": 38, "qatconfig": 38, "quantizationmethod": 39, "opquantizationconfig": 39, "attributequantizationconfig": 39, "quantizationconfigopt": 39, "targetplatformmodel": 39, "operatorsset": 39, "fuse": 39, "operatorsetconcat": 39, "operationstolay": 39, "operationssettolay": 39, "layerfilterparam": 39, "basekerastrainablequant": 40, "basepytorchtrainablequant": 40, "trainablequantizerweightsconfig": 40, "trainablequantizeractivationconfig": 40, "visual": 42, "within": 42, "tensorboard": 42, "cosin": 42, "similar": 42, "comparison": 42, "mix": 42, "precis": 42, "bit": 42, "width": 42, "compress": 43, "toolkit": 43, "user": 43, "guid": 43, "overview": 43, "instal": 43, "support": 43, "featur": 43, "quickstart": 43, "document": 43, "technic": 43, "constraint": 43, "refer": 43}, "envversion": {"sphinx.domains.c": 2, "sphinx.domains.changeset": 1, "sphinx.domains.citation": 1, "sphinx.domains.cpp": 6, "sphinx.domains.index": 1, "sphinx.domains.javascript": 2, "sphinx.domains.math": 2, "sphinx.domains.python": 3, "sphinx.domains.rst": 2, "sphinx.domains.std": 2, "sphinx": 56}}) \ No newline at end of file +Search.setIndex({"docnames": ["api/api_docs/classes/BitWidthConfig", "api/api_docs/classes/DataGenerationConfig", "api/api_docs/classes/DefaultDict", "api/api_docs/classes/FrameworkInfo", "api/api_docs/classes/GradientPTQConfig", "api/api_docs/classes/MixedPrecisionQuantizationConfig", "api/api_docs/classes/MpDistanceWeighting", "api/api_docs/classes/PruningConfig", "api/api_docs/classes/PruningInfo", "api/api_docs/classes/QuantizationConfig", "api/api_docs/classes/QuantizationErrorMethod", "api/api_docs/classes/ResourceUtilization", "api/api_docs/classes/XQuantConfig", "api/api_docs/index", "api/api_docs/methods/get_keras_data_generation_config", "api/api_docs/methods/get_keras_gptq_config", "api/api_docs/methods/get_pytorch_data_generation_config", "api/api_docs/methods/get_pytroch_gptq_config", "api/api_docs/methods/get_target_platform_capabilities", "api/api_docs/methods/keras_data_generation_experimental", "api/api_docs/methods/keras_gradient_post_training_quantization", "api/api_docs/methods/keras_kpi_data", "api/api_docs/methods/keras_load_quantizad_model", "api/api_docs/methods/keras_post_training_quantization", "api/api_docs/methods/keras_pruning_experimental", "api/api_docs/methods/keras_quantization_aware_training_finalize_experimental", "api/api_docs/methods/keras_quantization_aware_training_init_experimental", "api/api_docs/methods/pytorch_data_generation_experimental", "api/api_docs/methods/pytorch_gradient_post_training_quantization", "api/api_docs/methods/pytorch_kpi_data", "api/api_docs/methods/pytorch_post_training_quantization", "api/api_docs/methods/pytorch_pruning_experimental", "api/api_docs/methods/pytorch_quantization_aware_training_finalize_experimental", "api/api_docs/methods/pytorch_quantization_aware_training_init_experimental", "api/api_docs/methods/set_logger_path", "api/api_docs/methods/xquant_report_keras_experimental", "api/api_docs/methods/xquant_report_pytorch_experimental", "api/api_docs/modules/core_config", "api/api_docs/modules/debug_config", "api/api_docs/modules/exporter", "api/api_docs/modules/layer_filters", "api/api_docs/modules/network_editor", "api/api_docs/modules/qat_config", "api/api_docs/modules/target_platform", "api/api_docs/modules/trainable_infrastructure", "api/api_docs/notes/tpc_note", "guidelines/visualization", "index"], "filenames": ["api/api_docs/classes/BitWidthConfig.rst", "api/api_docs/classes/DataGenerationConfig.rst", "api/api_docs/classes/DefaultDict.rst", "api/api_docs/classes/FrameworkInfo.rst", "api/api_docs/classes/GradientPTQConfig.rst", "api/api_docs/classes/MixedPrecisionQuantizationConfig.rst", "api/api_docs/classes/MpDistanceWeighting.rst", "api/api_docs/classes/PruningConfig.rst", "api/api_docs/classes/PruningInfo.rst", "api/api_docs/classes/QuantizationConfig.rst", "api/api_docs/classes/QuantizationErrorMethod.rst", "api/api_docs/classes/ResourceUtilization.rst", "api/api_docs/classes/XQuantConfig.rst", "api/api_docs/index.rst", "api/api_docs/methods/get_keras_data_generation_config.rst", "api/api_docs/methods/get_keras_gptq_config.rst", "api/api_docs/methods/get_pytorch_data_generation_config.rst", "api/api_docs/methods/get_pytroch_gptq_config.rst", "api/api_docs/methods/get_target_platform_capabilities.rst", "api/api_docs/methods/keras_data_generation_experimental.rst", "api/api_docs/methods/keras_gradient_post_training_quantization.rst", "api/api_docs/methods/keras_kpi_data.rst", "api/api_docs/methods/keras_load_quantizad_model.rst", "api/api_docs/methods/keras_post_training_quantization.rst", "api/api_docs/methods/keras_pruning_experimental.rst", "api/api_docs/methods/keras_quantization_aware_training_finalize_experimental.rst", "api/api_docs/methods/keras_quantization_aware_training_init_experimental.rst", "api/api_docs/methods/pytorch_data_generation_experimental.rst", "api/api_docs/methods/pytorch_gradient_post_training_quantization.rst", "api/api_docs/methods/pytorch_kpi_data.rst", "api/api_docs/methods/pytorch_post_training_quantization.rst", "api/api_docs/methods/pytorch_pruning_experimental.rst", "api/api_docs/methods/pytorch_quantization_aware_training_finalize_experimental.rst", "api/api_docs/methods/pytorch_quantization_aware_training_init_experimental.rst", "api/api_docs/methods/set_logger_path.rst", "api/api_docs/methods/xquant_report_keras_experimental.rst", "api/api_docs/methods/xquant_report_pytorch_experimental.rst", "api/api_docs/modules/core_config.rst", "api/api_docs/modules/debug_config.rst", "api/api_docs/modules/exporter.rst", "api/api_docs/modules/layer_filters.rst", "api/api_docs/modules/network_editor.rst", "api/api_docs/modules/qat_config.rst", "api/api_docs/modules/target_platform.rst", "api/api_docs/modules/trainable_infrastructure.rst", "api/api_docs/notes/tpc_note.rst", "guidelines/visualization.rst", "index.rst"], "titles": ["BitWidthConfig", "Data Generation Configuration", "DefaultDict Class", "FrameworkInfo Class", "GradientPTQConfig Class", "MixedPrecisionQuantizationConfig", "MpDistanceWeighting", "Pruning Configuration", "Pruning Information", "QuantizationConfig", "QuantizationErrorMethod", "ResourceUtilization", "XQuant Configuration", "API Docs", "Get DataGenerationConfig for Keras Models", "Get GradientPTQConfig for Keras Models", "Get DataGenerationConfig for Pytorch Models", "Get GradientPTQConfig for Pytorch Models", "Get TargetPlatformCapabilities", "Keras Data Generation", "Keras Gradient Based Post Training Quantization", "Get Resource Utilization information for Keras Models", "Load Quantized Keras Model", "Keras Post Training Quantization", "Keras Structured Pruning", "Keras Quantization Aware Training Model Finalize", "Keras Quantization Aware Training Model Init", "Pytorch Data Generation", "Pytorch Gradient Based Post Training Quantization", "Get Resource Utilization information for PyTorch Models", "Pytorch Post Training Quantization", "Pytorch Structured Pruning", "PyTorch Quantization Aware Training Model Finalize", "PyTorch Quantization Aware Training Model Init", "Enable a Logger", "XQuant Report Keras", "XQuant Report Pytorch", "CoreConfig", "debug_config Module", "exporter Module", "Layer Attributes Filters", "network_editor Module", "qat_config Module", "target_platform Module", "trainable_infrastructure Module", "<no title>", "Visualization within TensorBoard", "Model Compression Toolkit User Guide"], "terms": {"class": [0, 1, 5, 6, 7, 8, 9, 10, 11, 12, 13, 22, 37, 38, 39, 40, 41, 42, 43, 44], "model_compression_toolkit": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 46], "core": [0, 3, 5, 6, 9, 10, 11, 20, 21, 23, 24, 25, 26, 28, 29, 31, 32, 33, 37, 38, 41], "manual_activation_bit_width_selection_list": 0, "none": [0, 1, 2, 4, 5, 12, 15, 17, 18, 20, 22, 23, 26, 28, 30, 33, 37, 39, 41, 42, 43, 44], "manag": 0, "manual": [0, 13, 37], "bit": [0, 5, 11, 13, 20, 23, 25, 26, 33, 37, 39, 41, 43, 44, 47], "width": [0, 5, 13, 20, 23, 26, 27, 33, 37, 43, 47], "configur": [0, 4, 5, 9, 11, 13, 14, 15, 16, 17, 19, 20, 23, 24, 25, 26, 27, 28, 30, 31, 32, 33, 35, 36, 37, 38, 40, 41, 42, 43, 44, 47], "A": [0, 3, 4, 5, 8, 13, 15, 17, 18, 20, 21, 22, 23, 24, 25, 26, 28, 29, 30, 31, 32, 33, 35, 36, 37, 38, 41, 42, 43, 47], "list": [0, 1, 3, 4, 5, 14, 15, 16, 17, 19, 27, 38, 41, 43, 47], "object": [0, 3, 4, 5, 7, 11, 14, 15, 16, 17, 18, 20, 21, 22, 23, 25, 26, 28, 29, 30, 33, 39, 41, 43, 44], "defin": [0, 4, 5, 6, 15, 17, 19, 20, 23, 24, 25, 26, 27, 28, 30, 31, 43, 44], "type": [0, 1, 2, 4, 5, 7, 8, 14, 15, 16, 17, 18, 19, 20, 21, 24, 25, 27, 29, 31, 35, 36, 39, 41, 43], "common": [0, 12], "quantiz": [0, 3, 4, 5, 9, 10, 11, 12, 13, 15, 17, 19, 21, 27, 29, 35, 36, 37, 38, 41, 42, 43, 44, 46, 47], "bit_width_config": [0, 37], "filter": [0, 1, 7, 43], "bit_width": 0, "encapsul": 0, "select": [0, 3, 7, 9, 10, 13, 37, 39, 42, 43, 44], "specif": [0, 3, 24, 31, 41, 46], "The": [0, 1, 3, 4, 5, 6, 7, 8, 9, 11, 12, 14, 15, 16, 17, 19, 20, 23, 24, 25, 26, 27, 28, 30, 31, 33, 35, 36, 39, 41, 43, 44, 46], "us": [0, 1, 2, 3, 4, 5, 6, 7, 9, 10, 11, 13, 14, 15, 16, 17, 18, 19, 20, 21, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 35, 36, 37, 38, 40, 41, 42, 43, 44, 45, 46, 47], "node": [0, 26, 33, 41, 44, 46], "manipul": [0, 1], "basenodematch": 0, "appli": [0, 1, 13, 39, 40, 41, 43], "int": [0, 1, 4, 5, 7, 9, 14, 15, 16, 17, 19, 27, 43, 44], "data_gener": [1, 14, 16, 19, 27], "datagenerationconfig": [1, 13, 19, 27], "n_iter": [1, 14, 16, 19, 27], "optim": [1, 3, 4, 9, 11, 13, 14, 15, 16, 17, 18, 20, 21, 23, 26, 28, 29, 30, 33, 37, 43, 44, 45, 47], "data_gen_batch_s": [1, 14, 16, 19, 27], "initial_lr": [1, 14, 16], "output_loss_multipli": [1, 14, 16], "image_granular": [1, 14, 16], "allimag": [1, 16], "scheduler_typ": [1, 14, 16], "bn_alignment_loss_typ": [1, 14, 16], "output_loss_typ": [1, 14, 16], "data_init_typ": [1, 14, 16], "layer_weighting_typ": [1, 14, 16], "image_pipeline_typ": [1, 14, 16], "image_normalization_typ": [1, 14, 16], "extra_pixel": [1, 14, 16], "0": [1, 3, 4, 5, 8, 9, 14, 15, 16, 20, 23, 24, 25, 26, 31, 39, 44], "bn_layer_typ": [1, 14, 16], "last_layer_typ": [1, 16], "image_clip": [1, 14, 16], "true": [1, 4, 5, 9, 15, 16, 17, 22, 32, 33, 44], "initi": [1, 2, 4, 7, 12, 14, 16, 26, 33, 43, 44], "paramet": [1, 2, 3, 4, 5, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44], "number": [1, 4, 5, 7, 14, 15, 16, 17, 19, 20, 23, 24, 26, 27, 28, 30, 31, 33, 43, 44], "iter": [1, 14, 16, 19, 20, 23, 26, 27, 28, 30, 33], "ani": [1, 2, 3, 4, 35, 36, 39, 40, 43, 44], "batch": [1, 4, 5, 14, 15, 16, 17, 19, 20, 23, 26, 27, 28, 30, 33], "size": [1, 4, 5, 14, 15, 16, 17, 19, 20, 23, 25, 26, 27, 33, 39, 42], "float": [1, 4, 5, 9, 14, 15, 16, 17, 20, 26, 28, 30, 33, 35, 36, 39, 43, 44, 46], "learn": [1, 14, 15, 16, 42], "rate": [1, 14, 15, 16], "multipli": [1, 14, 16], "output": [1, 3, 6, 9, 14, 16, 19, 20, 23, 26, 27, 28, 30, 32, 33, 43, 46, 47], "loss": [1, 4, 14, 15, 16, 17, 20, 24, 28, 30, 31], "granular": [1, 14, 16], "imag": [1, 5, 14, 16, 19, 20, 23, 26, 27, 28, 30, 33, 46], "default": [1, 2, 4, 7, 14, 15, 16, 18, 20, 23, 24, 28, 30, 31, 37, 39, 42, 43, 46], "schedul": [1, 14, 16, 38], "batchnorm": [1, 14, 16, 19, 20, 23, 26, 28, 30, 33], "align": [1, 14, 16], "layer": [1, 3, 5, 6, 8, 14, 16, 19, 20, 23, 24, 25, 26, 28, 30, 31, 32, 33, 38, 39, 41, 43, 44, 46], "weight": [1, 3, 4, 5, 6, 9, 11, 13, 14, 15, 16, 17, 20, 21, 24, 26, 28, 29, 30, 31, 32, 33, 39, 41, 42, 43, 44, 46], "pipelin": [1, 14, 16], "normal": [1, 4, 5, 14, 16], "union": [1, 14, 16, 19, 27, 43], "tupl": [1, 3, 14, 16, 19, 20, 24, 27, 31, 41, 43], "extra": [1, 14, 16], "pixel": [1, 14, 16], "add": [1, 3, 12, 14, 16, 22, 43, 44], "input": [1, 5, 9, 14, 16, 20, 23, 26, 28, 30, 33, 43], "bool": [1, 4, 5, 9, 14, 15, 16, 17, 38, 43, 44], "flag": 1, "enabl": [1, 5, 9, 13, 38, 44, 47], "clip": [1, 14, 16], "valu": [1, 2, 3, 4, 5, 6, 7, 9, 10, 11, 12, 20, 23, 24, 25, 26, 31, 39, 40, 41, 42, 43], "an": [1, 2, 3, 4, 8, 13, 20, 23, 26, 33, 35, 36, 39, 40, 41, 42, 43, 44, 47], "enum": [1, 3, 4, 5, 6, 7, 10, 42], "choos": [1, 4, 39], "depend": [1, 20, 23, 26, 28, 30, 33], "when": [1, 2, 3, 5, 7, 10, 11, 13, 15, 17, 20, 23, 25, 26, 38, 40, 42, 43, 44, 46], "imagewis": 1, "batchwis": [1, 14], "reduce_on_plateau": [1, 14], "reduceonplateau": 1, "step": [1, 42], "l2_squar": [1, 14, 16], "l2": 1, "squar": [1, 10], "No": 1, "negative_min_max_diff": [1, 16], "mean": [1, 4, 10, 46], "neg": [1, 9], "min": [1, 3, 9, 10, 20, 23, 26, 28, 30, 33, 46], "max": [1, 3, 9, 10, 20, 21, 23, 26, 28, 29, 30, 33, 46], "differ": [1, 5, 9, 13, 20, 23, 25, 26, 39, 43, 46], "inverse_min_max_diff": 1, "1": [1, 3, 4, 5, 8, 17, 19, 20, 21, 23, 24, 25, 27, 28, 29, 30, 31, 32, 39, 47], "regularized_min_max_diff": [1, 14], "regular": [1, 4, 15, 17], "gaussian": [1, 14, 16], "divers": 1, "averag": [1, 4, 5, 6, 14, 15, 16, 17], "same": [1, 39, 43], "per": [1, 3, 20, 23, 26, 33, 43, 44, 46], "first_layer_multipli": 1, "first": [1, 20, 23, 26, 28, 30, 33, 39, 46], "all": [1, 3, 4, 6, 9, 41, 44, 46], "other": [1, 15, 17], "grad": 1, "gradient": [1, 13, 30, 47], "base": [1, 4, 5, 10, 13, 15, 17, 19, 24, 27, 30, 31, 43, 44, 47], "smoothing_and_augment": [1, 14, 16], "smooth": [1, 42], "crop": 1, "flip": 1, "ident": 1, "do": [1, 46], "transform": [1, 20, 23, 26, 28, 30, 33], "torchvis": [1, 16, 28, 29, 30, 31, 32, 33, 39], "keras_appl": [1, 14], "imagenet": 1, "no_norm": 1, "known_dict": 2, "default_valu": 2, "dictionari": [2, 3, 4, 12, 25, 26, 35, 36, 39, 41, 42, 44], "It": [2, 12, 43, 44], "wrap": [2, 3, 9, 22, 26, 33, 40, 43, 44], "given": [2, 11, 20, 21, 23, 26, 28, 29, 30, 33], "return": [2, 4, 5, 8, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 35, 36, 39], "its": [2, 3, 9, 22, 24, 31, 40, 43, 46], "request": 2, "If": [2, 3, 5, 20, 23, 25, 26, 28, 30, 37, 39, 40, 43], "kei": [2, 6, 12, 24, 31, 40], "present": [2, 46], "pass": [2, 3, 9, 15, 17, 20, 23, 24, 25, 26, 28, 30, 31, 32, 33, 41], "gener": [2, 12, 13, 14, 16, 20, 21, 23, 24, 25, 26, 28, 29, 30, 31, 32, 33, 35, 36, 43, 46, 47], "provid": [2, 5, 19, 24, 27, 31, 39, 43, 44, 46], "empti": 2, "get": [2, 3, 4, 5, 13, 20, 23, 25, 26, 28, 30, 32, 33, 43, 46], "inner": 2, "default_factori": 2, "exist": [2, 41], "wa": [2, 39], "follow": [3, 4, 44, 46], "api": [3, 4, 23, 26, 33], "can": [3, 4, 6, 9, 13, 15, 17, 18, 19, 21, 24, 27, 29, 31, 38, 39, 41, 43, 44, 46, 47], "mct": [3, 9, 13, 15, 17, 18, 19, 20, 21, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 37, 38, 39, 41, 43, 44, 45, 46, 47], "framework": [3, 18, 43, 44], "relat": [3, 8, 13, 43], "inform": [3, 4, 13, 15, 17, 18, 20, 23, 24, 26, 28, 30, 31, 33, 38, 43, 44, 45], "network": [3, 7, 9, 32, 37, 38, 41, 46, 47], "activation_quantizer_map": 3, "kernel_channels_map": 3, "activation_min_max_map": 3, "layer_min_max_map": 3, "kernel_ops_attributes_map": 3, "out_channel_axis_map": 3, "about": [3, 4, 8, 13, 15, 17, 20, 23, 25, 26, 39, 43, 44], "librari": [3, 9], "need": [3, 20, 23, 26, 28, 30, 33, 39, 40, 44], "model": [3, 4, 5, 8, 9, 11, 12, 13, 18, 19, 20, 23, 24, 27, 28, 30, 31, 35, 36, 37, 41, 42, 43, 44, 46], "hold": [3, 37, 40, 43], "how": [3, 7, 20, 21, 23, 26, 28, 30, 33, 39, 44, 47], "thei": 3, "should": [3, 4, 7, 9, 11, 15, 17, 20, 21, 23, 24, 25, 26, 28, 30, 31, 33, 39, 43, 46], "multipl": [3, 5, 43], "map": [3, 43], "kernel": [3, 11, 20, 23, 25, 26, 41, 44], "channel": [3, 7, 8, 9, 13, 24, 31, 43, 44, 46], "indic": [3, 8, 24, 31, 43], "etc": [3, 11, 13, 20, 23, 26, 28, 30, 33, 46], "ar": [3, 4, 12, 18, 20, 23, 24, 26, 28, 30, 31, 33, 39, 43, 44, 45, 46], "divid": 3, "three": 3, "group": [3, 7, 24, 31, 43], "kernel_op": 3, "have": [3, 39, 40, 46], "coeffici": [3, 11, 20, 23, 25, 26, 28, 30, 43, 44], "e": [3, 20, 23, 26, 28, 30, 33, 47], "g": [3, 20, 23, 26, 28, 30, 33], "conv2d": [3, 11, 19, 20, 23, 25, 26, 27, 41, 43], "dens": [3, 19], "activation_op": 3, "relu": [3, 9], "no_quantization_op": 3, "reshap": [3, 19], "transpos": 3, "dict": [3, 4, 8, 12, 35, 36, 39, 43, 44], "quantizationmethod": [3, 44], "callabl": [3, 4, 5, 12, 15, 17, 20, 21, 23, 24, 26, 28, 29, 30, 31, 33, 35, 36, 39, 40], "from": [3, 4, 9, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 39, 41, 43, 44, 45, 46, 47], "function": [3, 4, 5, 6, 13, 14, 15, 16, 17, 19, 22, 24, 27, 31, 41, 43, 44], "defaultdict": [3, 13, 43], "out": [3, 7], "str": [3, 12, 35, 36, 39, 40, 43], "activ": [3, 9, 11, 20, 21, 23, 26, 28, 29, 30, 33, 39, 41, 42, 43, 44, 46], "oper": [3, 11, 38, 40, 43], "attirbut": 3, "s": [3, 7, 9, 11, 20, 23, 24, 25, 26, 28, 30, 31, 33, 39, 40, 41, 43, 44, 47], "comput": [3, 4, 5, 6, 10, 12, 13, 15, 17, 21, 29, 35, 38, 46], "statist": [3, 20, 23, 26, 28, 30, 33, 46], "exampl": [3, 4, 9, 11, 15, 17, 19, 20, 21, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 41, 43, 44, 47], "kera": [3, 4, 11, 13, 41, 44, 47], "we": [3, 19, 20, 23, 24, 26, 27, 31, 33, 39, 41, 43, 44, 46], "want": 3, "onli": [3, 4, 5, 6, 7, 11, 20, 23, 25, 26, 39], "set": [3, 6, 12, 13, 15, 19, 20, 23, 24, 25, 26, 27, 28, 30, 31, 33, 34, 35, 36, 41, 43, 44, 46], "know": 3, "3": [3, 15, 17, 19, 20, 21, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 39, 44], "2": [3, 9, 15, 19, 27, 43, 44, 47], "respectivli": 3, "import": [3, 7, 8, 9, 13, 15, 17, 19, 20, 21, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 39, 41, 44, 46], "tensorflow": [3, 13, 15, 18, 19, 20, 21, 23, 24, 25, 26, 39, 41, 43, 47], "tf": [3, 15, 19, 22, 25, 26], "Then": [3, 20, 23, 26, 28, 30, 33, 41, 46], "creat": [3, 4, 9, 13, 14, 15, 16, 17, 19, 20, 21, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 39, 40, 41], "rang": [3, 20, 23, 26, 28, 30, 33], "advanc": 3, "save": [3, 12, 26, 34, 39, 44], "collect": [3, 20, 23, 26, 28, 30, 33, 35, 36, 46], "time": [3, 7, 44], "For": [3, 9, 18, 19, 20, 23, 25, 26, 27, 33, 39, 43, 44, 45, 46, 47], "softmax": [3, 9], "format": [3, 13], "index": [3, 13], "axi": [3, 44], "nhwc": 3, "last": [3, 6], "nchw": 3, "instanc": [4, 13, 15, 17, 41, 43, 46], "which": [4, 7, 9, 39, 40, 41, 43, 44], "post": [4, 13, 24, 26, 31, 33, 47], "train": [4, 13, 42, 44, 47], "knowledg": [4, 47], "distil": [4, 47], "teacher": 4, "student": 4, "gptq": [4, 15, 17, 20, 28], "n_epoch": [4, 15, 17, 20], "optimizer_rest": [4, 15, 17], "log_funct": [4, 15, 17], "train_bia": 4, "rounding_typ": 4, "softquant": 4, "use_hessian_based_weight": [4, 15, 17], "optimizer_quantization_paramet": 4, "optimizer_bia": 4, "regularization_factor": [4, 15, 17], "reg_default": [4, 15, 17], "hessian_weights_config": 4, "gptq_quantizer_params_overrid": 4, "gradientptq": [4, 13], "repres": [4, 5, 11, 15, 17, 20, 23, 24, 25, 26, 28, 30, 31, 32, 33, 35, 36, 39, 41, 43, 46], "dataset": [4, 15, 17, 20, 21, 23, 24, 25, 26, 28, 29, 30, 31, 32, 33, 35, 36, 39, 46], "epoch": [4, 15, 17], "bia": [4, 9, 11, 15, 17, 20, 23, 25, 26], "accept": [4, 15, 17, 43], "6": [4, 27], "tensor": [4, 5, 12, 15, 17, 19, 21, 27, 29, 43, 44, 46, 47], "1st": [4, 15, 17], "2nd": [4, 15, 17], "3rd": [4, 15, 17], "4th": [4, 15, 17], "5th": 4, "6th": 4, "std": 4, "accordingli": [4, 43], "see": [4, 47], "multiple_tensors_mse_loss": [4, 17], "log": [4, 12, 13, 15, 17, 34, 46], "process": [4, 5, 9, 13, 14, 15, 16, 17, 18, 19, 24, 27, 31, 37, 38, 41, 42, 43, 45, 46], "whether": [4, 5, 8, 9, 14, 15, 16, 17, 22, 38, 39, 43, 44], "updat": 4, "dure": [4, 9, 13, 14, 15, 16, 17, 18, 35, 36, 39, 41, 43, 44, 45, 46], "round": 4, "hessian": [4, 5, 7, 10, 15, 17, 24, 31, 47], "overrid": [4, 42], "rest": 4, "point": [4, 5, 15, 17, 20, 28, 30, 35, 36, 43, 46], "factor": [4, 5, 10, 15, 17], "includ": [4, 8, 11, 20, 23, 26, 28, 30, 33, 42], "necessari": [4, 39, 44], "argument": [4, 6, 43], "run": [4, 15, 17, 39, 46], "score": [4, 5, 7, 8, 9, 10, 24, 31], "instanti": [4, 9, 42], "hessians_num_sampl": 4, "gptq_hessian_num_sampl": 4, "norm_scor": [4, 5], "log_norm": 4, "scale_log_norm": 4, "fals": [4, 5, 9, 14, 38, 43], "hessian_batch_s": [4, 5, 15, 17], "act_hessian_default_batch_s": [4, 5, 15, 17], "metric": [4, 5, 6, 7, 11, 12, 13, 35, 36], "gptqhessianweightsconfig": 4, "sampl": [4, 46], "between": [4, 5, 9, 12, 20, 28, 30, 43, 46], "scale": [4, 5, 9, 43], "final": [4, 5, 13, 19, 27, 41, 46, 47], "vector": [4, 46], "method": [4, 5, 6, 7, 9, 10, 13, 24, 31, 39, 41, 42, 43, 44], "ste": [4, 42], "straight": [4, 42], "through": [4, 19, 24, 27, 42], "estim": [4, 42], "mix": [5, 6, 11, 13, 20, 21, 23, 25, 26, 28, 29, 30, 33, 37, 43, 47], "precis": [5, 6, 11, 13, 20, 21, 23, 24, 25, 26, 28, 29, 30, 31, 33, 37, 43, 47], "compute_distance_fn": 5, "distance_weighting_method": 5, "mpdistanceweight": [5, 13], "avg": [5, 6], "num_of_imag": [5, 20, 23], "mp_default_num_sampl": 5, "configuration_overwrit": 5, "num_interest_points_factor": 5, "use_hessian_based_scor": 5, "refine_mp_solut": 5, "metric_normalization_threshold": 5, "1e10": 5, "distanc": [5, 6, 13], "two": [5, 12, 20, 23, 26, 28, 30, 33, 39, 43, 46], "pre": 5, "each": [5, 7, 8, 20, 23, 24, 26, 28, 30, 31, 33, 41, 43, 44, 46], "among": 5, "sensit": [5, 7, 24, 31], "evalu": [5, 35, 36], "compar": [5, 20, 28, 30, 46], "integ": [5, 39, 43], "overwrit": 5, "predefin": [5, 7], "one": [5, 9, 40, 46], "zero": [5, 43], "percentag": 5, "reduc": [5, 24, 31], "interest": 5, "calcul": [5, 7, 13, 20, 21, 23, 24, 26, 28, 29, 30, 31, 33], "try": 5, "improv": [5, 24, 31], "greedi": [5, 7], "algorithm": 5, "search": [5, 9, 11, 13, 20, 23, 26, 28, 30, 33], "increas": 5, "threshold": [5, 9, 10, 20, 23, 26, 28, 30, 33, 43, 44], "check": [5, 39, 40, 41], "In": [5, 19, 20, 23, 26, 27, 28, 30, 33, 39, 40], "case": 5, "larger": 5, "than": [5, 40], "thi": [5, 8, 10, 11, 13, 19, 20, 22, 23, 24, 25, 26, 27, 28, 30, 31, 33, 39, 43, 44, 47], "prevent": 5, "numer": 5, "issu": [5, 39], "call": [6, 21, 29, 43, 46], "take": [6, 23, 26, 33, 47], "last_lay": 6, "pruningconfig": [7, 13, 24, 31], "num_score_approxim": [7, 24, 31], "pruning_num_score_approxim": 7, "importance_metr": 7, "lfh": [7, 24, 31], "channels_filtering_strategi": 7, "specifi": [7, 14, 16, 19, 22, 24, 27, 31, 39], "neural": [7, 47], "approxim": [7, 24, 31], "perform": [7, 11, 19, 24, 27, 31], "strategi": [7, 24, 31], "constant": [7, 41, 44], "label": [7, 24, 31, 43, 47], "free": [7, 19, 24, 27, 31, 47], "approach": 7, "determin": [7, 24, 31], "info": [7, 34], "measur": [7, 11, 12, 46], "least": 7, "up": [7, 19, 27, 43, 46], "allow": [7, 12, 19, 27, 39], "resourc": [7, 11, 13, 20, 23, 24, 25, 26, 31, 32, 33, 46], "util": [7, 11, 13, 20, 23, 24, 25, 26, 31, 32, 33, 44], "limit": [7, 20, 23, 25, 26, 28, 30, 33], "now": [7, 18, 33, 39, 43, 44, 45, 46], "weights_memori": [7, 11, 20, 23, 24, 26, 31, 33], "consid": [7, 14, 16, 24, 31, 43], "pruninginfo": [8, 13, 24, 31], "pruning_mask": 8, "importance_scor": 8, "store": [8, 44], "mask": 8, "act": 8, "contain": [8, 13, 19, 20, 21, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 35, 36, 43, 44], "access": 8, "metadata": [8, 43], "basenod": 8, "np": [8, 11, 20, 21, 23, 24, 25, 26, 28, 29, 30, 31, 32, 33, 39], "ndarrai": 8, "arrai": 8, "where": [8, 12, 39, 41, 43, 46], "element": [8, 43], "correspond": 8, "neuron": 8, "ha": [8, 39, 40, 41], "been": 8, "kept": [8, 26, 33], "quantifi": [8, 46], "signific": 8, "properti": 8, "activation_error_method": 9, "quantizationerrormethod": [9, 13], "mse": [9, 10, 46], "weights_error_method": 9, "relu_bound_to_power_of_2": 9, "weights_bias_correct": 9, "weights_second_moment_correct": 9, "input_sc": 9, "softmax_shift": 9, "shift_negative_activation_correct": 9, "activation_channel_equ": 9, "z_threshold": 9, "math": 9, "inf": [9, 11], "min_threshold": [9, 44], "l_p_valu": 9, "linear_collaps": 9, "residual_collaps": 9, "shift_negative_ratio": 9, "05": 9, "shift_negative_threshold_recalcul": 9, "shift_negative_params_search": 9, "concat_threshold_upd": 9, "accord": [9, 11, 13, 20, 21, 23, 24, 26, 28, 29, 30, 31, 33, 39, 40, 43], "power": [9, 20, 23, 26, 28, 30, 33, 43], "correct": 9, "second_mo": 9, "shift": 9, "equal": [9, 40], "z": 9, "outlier": 9, "remov": [9, 24, 31, 32], "minimum": [9, 44], "p": [9, 31], "l_p": 9, "norm": [9, 10, 46], "block_collaps": 9, "collaps": 9, "block": [9, 44, 46], "anoth": 9, "ratio": 9, "minim": [9, 10, 20, 24, 28, 30, 31], "non": [9, 43], "linear": [9, 27], "abov": 9, "occur": 9, "recomput": 9, "after": [9, 13, 20, 22, 23, 26, 33, 47], "One": [9, 46], "mai": [9, 19, 20, 23, 26, 27, 28, 30, 33, 40, 46], "done": [9, 46], "noclip": [9, 10], "qc": 9, "instans": 9, "keras_post_training_quant": [9, 13, 15, 23, 39, 41, 46], "error": 10, "nois": 10, "mae": [10, 46], "absolut": 10, "kl": [10, 46], "diverg": [10, 46], "make": 10, "signal": 10, "distribut": 10, "similar": [10, 12, 35, 36, 38, 47], "possibl": [10, 20, 23, 26, 33, 43, 46], "lp": 10, "hmse": 10, "more": [10, 18, 23, 24, 26, 31, 33, 39, 43, 45, 46], "valuabl": 10, "induc": 10, "prune": [11, 47], "activation_memori": 11, "total_memori": 11, "bop": 11, "memori": [11, 24, 31, 46], "byte": [11, 20, 23, 24, 26, 31, 33, 46], "note": [11, 20, 23, 25, 26], "affect": [11, 20, 23, 25, 26], "while": [11, 20, 23, 25, 26, 33, 43], "sum": [11, 21, 24, 29, 31], "total": [11, 21, 29], "xquant_config": [12, 35, 36], "xquantconfig": [12, 13, 35, 36], "report_dir": 12, "custom_similarity_metr": 12, "report": [12, 13], "dir": [12, 46], "explain": [12, 13, 35, 36, 44], "directori": [12, 13, 34], "custom": [12, 19, 22, 26, 27, 39], "name": [12, 18, 39, 41, 43, 46], "implement": [12, 44], "init": [13, 41, 47], "modul": [13, 27, 28, 29, 30, 31, 36], "pytorch_post_training_quant": [13, 17, 30, 39], "pytorch": [13, 18, 43, 44, 47], "pytorch_gradient_post_training_quant": [13, 28], "get_pytorch_gptq_config": [13, 17], "gradientptqconfig": [13, 20, 28], "keras_gradient_post_training_quant": [13, 20], "get_keras_gptq_config": [13, 15, 20], "option": [13, 22, 23, 24, 26, 28, 31, 33, 39, 43], "pytorch_quantization_aware_training_init_experiment": [13, 32, 33], "prepar": [13, 26, 33], "awar": [13, 42, 44, 47], "experiment": [13, 19, 27, 47], "pytorch_quantization_aware_training_finalize_experiment": [13, 32], "without": 13, "quantizewrapp": [13, 26, 32, 33], "keras_quantization_aware_training_init_experiment": [13, 25, 26], "keras_quantization_aware_training_finalize_experiment": [13, 25], "qat_config": [13, 26, 33], "coreconfig": [13, 20, 21, 23, 25, 26, 28, 29, 30, 32, 33], "entir": 13, "quantizationconfig": [13, 37], "mixedprecisionquantizationconfig": [13, 20, 21, 23, 25, 26, 37], "bitwidthconfig": [13, 37], "resourceutil": [13, 20, 21, 23, 24, 25, 26, 28, 29, 30, 31, 33], "network_editor": [13, 38], "modifi": [13, 41], "troubleshoot": 13, "pytorch_resource_utilization_data": [13, 29], "data": [13, 14, 16, 21, 24, 29, 31, 35, 36, 39, 43, 46, 47], "desir": [13, 20, 21, 23, 25, 26, 28, 29, 30, 33], "target": [13, 18, 20, 21, 23, 24, 25, 26, 29, 31, 32, 33], "keras_resource_utilization_data": [13, 21], "pytorch_data_generation_experiment": [13, 27], "get_pytorch_data_generation_config": [13, 16, 27], "load": [13, 25, 26, 39, 44], "keras_data_generation_experiment": [13, 19], "get_keras_data_generation_config": [13, 14, 19], "pytorch_pruning_experiment": [13, 31], "structur": [13, 47], "keras_pruning_experiment": [13, 24], "xquant_report_pytorch_experiment": [13, 36], "xquant_report_keras_experiment": [13, 35], "serial": 13, "abstract": [13, 44], "hardwar": [13, 18, 24, 31, 43, 44, 47], "orient": [13, 44], "tool": [13, 44], "logger": [13, 38, 46], "path": [13, 22, 34, 39, 46], "infer": [13, 18, 25, 32, 43, 44], "get_target_platform_cap": [13, 18, 43], "platform": [13, 18, 20, 23, 24, 25, 26, 29, 31], "targetplatformcap": [13, 20, 21, 23, 24, 26, 28, 29, 30, 31, 33], "page": 13, "document": [13, 23, 26, 33], "auto": 13, "sphinx": 13, "default_n_it": [14, 16], "adam": [14, 15, 17], "default_data_gen_b": [14, 16], "default_keras_initial_lr": 14, "default_keras_output_loss_multipli": 14, "schedulertyp": [14, 16], "batchnormalignemntlosstyp": [14, 16], "outputlosstyp": [14, 16], "datainittyp": [14, 16], "bnlayerweightingtyp": [14, 16], "imagegranular": [14, 16], "imagepipelinetyp": [14, 16], "imagenormalizationtyp": [14, 16], "default_keras_extra_pixel": 14, "learning_r": 15, "lr_default": [15, 17], "lr_rest_default": [15, 17], "gptqmultipletensorsloss": 15, "gradientptqconfigv2": [15, 17], "fine": [15, 17, 24, 25, 26, 31, 32, 33], "tune": [15, 17, 24, 25, 26, 31, 32], "optimizerv2": 15, "auxiliri": [15, 17], "variabl": [15, 17], "4": [15, 17, 19, 20, 23, 24, 26, 27, 28, 30, 31, 33], "5": [15, 17, 24, 31], "gptq_conf": [15, 17, 28], "nadam": 15, "order": [15, 17, 20, 23, 26, 33, 38, 39, 40], "radam": 16, "default_pytorch_initial_lr": 16, "default_pytorch_output_loss_multipli": 16, "reduce_on_plateau_with_reset": 16, "default_pytorch_extra_pixel": 16, "default_pytorch_bn_layer_typ": 16, "default_pytorch_last_layer_typ": 16, "torch": [17, 27, 36, 39, 47], "lr": 17, "dummi": 17, "param": [17, 38, 41, 44], "fw_name": 18, "target_platform_nam": 18, "target_platform_vers": 18, "support": [18, 39], "both": [18, 20, 23, 28, 30, 32, 44, 46], "them": [18, 43, 46], "imx500": [18, 39, 43], "tflite": [18, 39, 43], "qnnpack": [18, 43], "capabl": [18, 24, 29, 31], "version": [18, 19, 27, 43, 47], "attach": [18, 43], "some": [18, 19, 27, 39, 43, 45, 46], "field": [18, 40, 43, 45], "opquantizationconfig": [18, 45], "ignor": [18, 43, 45], "quantization_preserv": [18, 43, 45], "fixed_scal": [18, 43, 45], "fixed_zero_point": [18, 43, 45], "futur": [18, 19, 27, 43, 45], "n_imag": [19, 27], "output_image_s": [19, 27], "data_generation_config": [19, 27], "ll": [19, 27], "walk": [19, 27], "simpl": [19, 27], "involv": [19, 24, 27, 31], "start": [19, 27, 39, 44, 47], "compress": [19, 24, 27, 28, 31], "toolkit": [19, 27, 28], "sequenti": [19, 27], "flatten": [19, 27], "next": [19, 27, 39, 40], "input_shap": 19, "8": [19, 20, 23, 25, 26, 27, 39, 44], "10": [19, 20, 23, 26, 27, 28, 30, 33], "simplic": [19, 27], "config": [19, 20, 23, 24, 25, 26, 27, 28, 31, 32, 33, 37, 41, 43, 44], "notic": [19, 24, 27, 31, 39], "chang": [19, 27, 39, 41, 46], "generated_imag": [19, 27], "variou": [19, 27, 46], "purpos": [19, 27, 38], "in_model": [20, 21, 23, 25, 26, 29, 32, 33], "representative_data_gen": [20, 21, 23, 24, 26, 28, 29, 30, 31, 33, 39], "gptq_config": [20, 28, 30], "gptq_representative_data_gen": [20, 28], "target_resource_util": [20, 23, 24, 26, 28, 30, 31, 33], "core_config": [20, 21, 23, 25, 26, 28, 29, 30, 32, 33], "target_platform_cap": [20, 21, 23, 24, 26, 28, 29, 30, 31, 33, 44], "default_keras_tpc": [20, 23, 24, 26], "symmetr": [20, 23, 26, 28, 30, 33, 42, 43, 44], "constraint": [20, 23, 24, 28, 30, 31], "sever": [20, 23, 26, 28, 30, 33, 46], "fold": [20, 23, 26, 28, 30, 33], "preced": [20, 23, 26, 28, 30, 33], "histogram": [20, 23, 26, 28, 30, 33, 46], "being": [20, 23, 26, 28, 30, 33, 43, 44], "ilp": [20, 23, 26, 33], "solver": [20, 23, 26, 33], "find": [20, 23, 26, 33], "maxim": [20, 23, 26, 33], "observ": [20, 28, 30, 43, 46], "calibr": [20, 21, 23, 26, 28, 29, 30, 33], "user": [20, 23, 25, 26, 28, 30, 32, 33], "handl": [20, 23, 26, 28, 30, 33], "applic": [20, 21, 23, 24, 25, 26, 39], "mobilenet": [20, 21], "random": [20, 21, 23, 24, 25, 26, 28, 29, 30, 31, 32, 33, 39], "requir": [20, 23, 26, 28, 30, 33, 44, 46], "num_calibration_batch": [20, 23, 26, 28, 30, 33], "numpi": [20, 21, 23, 24, 25, 26, 28, 29, 30, 31, 32, 33, 39], "def": [20, 21, 23, 24, 25, 26, 28, 29, 30, 31, 32, 33, 39], "repr_datagen": [20, 21, 23, 24, 25, 26, 28, 29, 30, 31, 32, 33], "_": [20, 23, 26, 28, 30, 33, 39], "yield": [20, 21, 23, 24, 25, 26, 28, 29, 30, 31, 32, 33, 39], "224": [20, 21, 23, 24, 25, 26, 28, 29, 30, 31, 32, 33, 39], "bitwidth": [20, 23, 25, 26], "candid": [20, 23, 25, 26, 41], "mixed_precision_config": [20, 21, 23, 25, 26, 37], "our": [20, 23, 25, 26, 33, 47], "ru": [20, 23, 25, 26], "count_param": [20, 23, 24, 25, 26], "75": [20, 23, 25, 26], "quantized_model": [20, 23, 25, 26, 32, 33, 35, 36], "quantization_info": [20, 23, 25, 26, 28, 30, 32, 33], "userinform": 20, "keras_default_tpc": 21, "build": [21, 29, 44, 47], "graph": [21, 29, 41, 46], "hw": 21, "ru_data": [21, 29], "keras_load_quantized_model": 22, "filepath": 22, "custom_object": [22, 25, 26], "compil": 22, "trainabl": [22, 25, 44], "file": [22, 25, 26, 34, 39], "addit": [22, 39], "boolean": 22, "saved_model": 22, "loadopt": 22, "savedmodel": 22, "ptq": [23, 30, 39], "mobilenet_v2": [23, 25, 26, 28, 29, 30, 32, 33, 39], "mobilenetv2": [23, 25, 26, 39, 46], "pleas": [23, 26, 33, 39, 47], "look": [23, 26, 33, 43, 47], "pruning_config": [24, 31], "meet": [24, 31], "simd": [24, 31, 43], "tpc": [24, 31, 43], "By": [24, 28, 30, 31, 39, 46], "assess": [24, 31], "togeth": [24, 31], "friendli": [24, 31, 47], "architectur": [24, 31], "analyz": [24, 31], "identifi": [24, 31], "impact": [24, 31], "must": [24, 31, 43], "retrain": [24, 31], "recov": [24, 31], "origin": [24, 35, 36, 46], "achiev": 24, "analysi": [24, 31], "standard": [24, 31, 42], "associ": [24, 31, 43], "resnet50": [24, 31, 39], "here": [24, 31, 39, 43, 47], "aim": [24, 31], "footprint": [24, 31], "50": [24, 31], "assum": [24, 31], "float32": [24, 31, 39], "thu": [24, 31, 46], "dense_nparam": [24, 31], "l": [24, 47], "higher": [24, 31], "also": [24, 31, 47], "extend": [24, 31], "durat": [24, 31], "pruned_model": [24, 31], "pruning_info": [24, 31], "qat": [25, 26, 32, 33, 42], "convert": [25, 32], "replac": 25, "trainablequant": 25, "inferablequant": [25, 32], "load_model": [25, 26], "model_fil": [25, 26], "qatconfig": [26, 33], "built": [26, 33, 44], "fake_qu": [26, 33], "onlin": [26, 33], "wrapper": [26, 32, 33, 44], "ker": 26, "hight": 27, "nn": [27, 36], "batchnorm2d": 27, "default_pytorch_tpc": [28, 30, 31, 33], "quantized_modul": [28, 30], "pytorch_default_tpc": 29, "in_modul": 30, "clibrat": 30, "default_pyotrch_tpc": 31, "resnet50_weight": 31, "imagenet1k_v1": 31, "numel": 31, "state_dict": 31, "pretrain": [32, 33], "keep": [32, 47], "readi": 32, "tun": 33, "set_log_fold": [34, 46], "folder": 34, "level": 34, "verbos": 34, "facade_xquant_report": [35, 36], "float_model": [35, 36, 39], "repr_dataset": [35, 36, 39], "validation_dataset": [35, 36], "valid": [35, 36, 44], "quantization_config": [37, 44], "debug_config": 37, "debugconfig": 37, "debug": [37, 38], "edit": [37, 38, 41], "analyze_similar": 38, "simulate_schedul": 38, "plot": [38, 46], "figur": [38, 46], "within": [38, 47], "tensorboard": [38, 47], "pinpoint": 38, "problemat": 38, "editrul": 38, "rule": [38, 41], "action": 38, "simul": 38, "behaviour": 38, "cut": 38, "detail": [39, 43], "refer": 39, "project": [39, 47], "github": [39, 47], "readm": 39, "you": [39, 46, 47], "question": 39, "open": [39, 46, 47], "repositori": 39, "fakely_qu": 39, "int8": 39, "mct_quantiz": 39, "via": [39, 47], "save_model_path": 39, "is_layer_exportable_fn": 39, "is_keras_layer_export": 39, "serialization_format": 39, "quantization_format": 39, "To": [39, 46], "demonstr": [39, 43], "quantized_exportable_model": 39, "lambda": 39, "dtype": 39, "avail": [39, 47], "tempfil": 39, "keras_file_path": 39, "mkstemp": 39, "torchscript": 39, "is_pytorch_layer_export": 39, "onnx_opset_vers": 39, "default_onnx_opset_vers": 39, "current": 39, "fake": 39, "trace": 39, "mandatori": 39, "quant": 39, "your": 39, "packag": [39, 44, 47], "so": [39, 43], "part": 39, "skip": 39, "plan": 39, "pip": [39, 47], "instal": 39, "q": 39, "onnxruntim": 39, "extens": 39, "let": 39, "There": [39, 46], "onnx_file_path": 39, "model_format_onnx_mctq": 39, "15": 39, "16": 39, "get_ort_session_opt": 39, "session": 39, "creation": 39, "slowli": 39, "suffer": 39, "longer": 39, "latenc": 39, "howev": 39, "ort": 39, "sess": 39, "inferencesess": 39, "cudaexecutionprovid": 39, "cpuexecutionprovid": 39, "_input_data": 39, "astyp": 39, "_model_output_nam": 39, "get_output": 39, "_model_input_nam": 39, "get_input": 39, "predict": 39, "layerfilterparam": 40, "target_platform": [40, 44], "attributefilt": [40, 43], "attr": 40, "op": [40, 43], "match": [40, 41, 43], "regard": 40, "eq": 40, "noteq": 40, "greater": 40, "greatereq": 40, "smaller": 40, "smallereq": 40, "attribut": [41, 43, 44], "9": 41, "er_list": 41, "nodetypefilt": 41, "changecandidatesweightsquantconfigattr": 41, "attr_nam": 41, "weights_n_bit": [41, 43, 44], "new": 41, "node_typ": 41, "nodenamefilt": 41, "node_nam": 41, "nodenamescopefilt": 41, "node_name_scop": 41, "string": 41, "changefinalweightsquantconfigattr": 41, "kwarg": [41, 43], "attr_valu": 41, "changefinalactivationquantconfigattr": 41, "changecandidatesactivationquantconfigattr": 41, "changequantizationparamfunct": 41, "activation_quantization_params_fn": 41, "weights_quantization_params_fn": 41, "changefinalweightsquantizationmethod": 41, "weights_quantization_method": [41, 43, 44], "changecandidatesweightsquantizationmethod": 41, "changecandidatesactivationquantizationmethod": 41, "activation_quantization_method": [41, 43, 44], "changequantizationmethod": 41, "poweroftwo": 42, "uniform": [42, 43, 44], "dqa": 42, "dnn": 42, "attent": 42, "introduc": 42, "lsq": 42, "http": [42, 47], "arxiv": [42, 47], "org": 42, "pdf": 42, "1902": 42, "08153": 42, "weight_training_method": 42, "activation_training_method": 42, "weight_quantizer_params_overrid": 42, "activation_quantizer_params_overrid": 42, "backend": 43, "address": 43, "shortli": 43, "diagram": 43, "main": [43, 46], "compon": [43, 44], "power_of_two": 43, "lut_pot_quant": 43, "lookup": 43, "tabl": 43, "lut_sym_quant": 43, "default_weight_attr_config": 43, "attr_weights_configs_map": 43, "activation_n_bit": [43, 44], "supported_input_activation_n_bit": 43, "enable_activation_quant": [43, 44], "simd_siz": 43, "signed": 43, "singl": 43, "instruct": 43, "fetch": 43, "simultan": 43, "float_bitwidth": 43, "weights_per_channel_threshold": [43, 44], "enable_weights_quant": [43, 44], "lut_values_bitwidth": 43, "quantization_config_list": 43, "base_config": 43, "gather": [43, 46], "fallback": 43, "manner": 43, "default_qco": 43, "add_metadata": 43, "default_tp_model": 43, "definit": 43, "pattern": 43, "combin": 43, "qc_option": 43, "uniqu": 43, "operator_groups_list": 43, "treat": 43, "henc": 43, "either": 43, "opset": 43, "concaten": 43, "similarli": 43, "place": 43, "like": 43, "op_sets_to_lay": 43, "op_set_nam": 43, "attr_map": 43, "filterlayerparam": 43, "condit": 43, "satisfi": 43, "keyword": 43, "usag": 43, "tp_model": 43, "infrastructur": 44, "propos": 44, "emul": 44, "holder": 44, "upon": 44, "learnabl": 44, "basic": 44, "get_config": 44, "from_config": 44, "weights_quantization_param": 44, "weights_channels_axi": 44, "weights_quantization_candid": 44, "activation_quantization_param": 44, "activation_quantization_candid": 44, "phase": 46, "displai": 46, "ui": 46, "writer": 46, "stage": 46, "launch": 46, "logdir": 46, "seen": 46, "under": 46, "tab": 46, "click": 46, "deeper": 46, "view": 46, "tag": 46, "show": 46, "These": 46, "scalar": 46, "were": 46, "wai": [46, 47], "mathemat": 46, "divis": 46, "dot": 46, "product": 46, "euclidean": 46, "along": 46, "pair": 46, "would": 46, "expect": 46, "mani": 46, "few": [46, 47], "As": 46, "20": 46, "insert": 46, "chosen": 46, "second": 46, "sourc": 47, "research": 47, "develop": 47, "engin": 47, "easili": 47, "state": 47, "art": 47, "work": 47, "soni": 47, "semiconductor": 47, "israel": 47, "git": 47, "clone": 47, "com": 47, "model_optim": 47, "python": 47, "setup": 47, "py": 47, "pypi": 47, "latest": 47, "stabl": 47, "releas": 47, "nightli": 47, "unstabl": 47, "visual": 47, "cosin": 47, "comparison": 47, "just": 47, "minut": 47, "visit": 47, "notebook": 47, "doesn": 47, "t": 47, "out1": 47, "out2": 47, "out3": 47, "habi": 47, "h": 47, "v": 47, "peretz": 47, "r": 47, "cohen": 47, "dikstein": 47, "dror": 47, "o": 47, "diamant": 47, "i": 47, "jen": 47, "netzer": 47, "2021": 47, "hptq": 47, "preprint": 47, "gordon": 47, "2023": 47, "eptq": 47, "enhanc": 47}, "objects": {"model_compression_toolkit": [[2, 0, 1, "", "DefaultDict"], [18, 3, 1, "", "get_target_platform_capabilities"], [22, 3, 1, "", "keras_load_quantized_model"], [34, 3, 1, "", "set_log_folder"]], "model_compression_toolkit.DefaultDict": [[2, 1, 1, "", "get"], [2, 1, 1, "", "keys"]], "model_compression_toolkit.core": [[0, 0, 1, "", "BitWidthConfig"], [3, 0, 1, "", "ChannelAxis"], [37, 0, 1, "", "CoreConfig"], [38, 0, 1, "", "DebugConfig"], [3, 0, 1, "", "FrameworkInfo"], [5, 0, 1, "", "MixedPrecisionQuantizationConfig"], [6, 0, 1, "", "MpDistanceWeighting"], [9, 0, 1, "", "QuantizationConfig"], [10, 0, 1, "", "QuantizationErrorMethod"], [11, 0, 1, "", "ResourceUtilization"], [21, 3, 1, "", "keras_resource_utilization_data"], [29, 3, 1, "", "pytorch_resource_utilization_data"]], "model_compression_toolkit.core.BitWidthConfig": [[0, 2, 1, "", "manual_activation_bit_width_selection_list"]], "model_compression_toolkit.core.common.quantization.bit_width_config": [[0, 0, 1, "", "ManualBitWidthSelection"]], "model_compression_toolkit.core.common.quantization.bit_width_config.ManualBitWidthSelection": [[0, 2, 1, "", "bit_width"], [0, 2, 1, "", "filter"]], "model_compression_toolkit.core.network_editor": [[41, 0, 1, "", "ChangeCandidatesActivationQuantConfigAttr"], [41, 0, 1, "", "ChangeCandidatesActivationQuantizationMethod"], [41, 0, 1, "", "ChangeCandidatesWeightsQuantConfigAttr"], [41, 0, 1, "", "ChangeCandidatesWeightsQuantizationMethod"], [41, 0, 1, "", "ChangeFinalActivationQuantConfigAttr"], [41, 0, 1, "", "ChangeFinalWeightsQuantConfigAttr"], [41, 0, 1, "", "ChangeFinalWeightsQuantizationMethod"], [41, 0, 1, "", "ChangeQuantizationParamFunction"], [41, 0, 1, "", "EditRule"], [41, 0, 1, "", "NodeNameFilter"], [41, 0, 1, "", "NodeNameScopeFilter"], [41, 0, 1, "", "NodeTypeFilter"]], "model_compression_toolkit.data_generation": [[1, 0, 1, "", "BNLayerWeightingType"], [1, 0, 1, "", "BatchNormAlignemntLossType"], [1, 0, 1, "", "DataGenerationConfig"], [1, 0, 1, "", "DataInitType"], [1, 0, 1, "", "ImageGranularity"], [1, 0, 1, "", "ImageNormalizationType"], [1, 0, 1, "", "ImagePipelineType"], [1, 0, 1, "", "OutputLossType"], [1, 0, 1, "", "SchedulerType"], [14, 3, 1, "", "get_keras_data_generation_config"], [16, 3, 1, "", "get_pytorch_data_generation_config"], [19, 3, 1, "", "keras_data_generation_experimental"], [27, 3, 1, "", "pytorch_data_generation_experimental"]], "model_compression_toolkit.exporter": [[39, 0, 1, "", "KerasExportSerializationFormat"], [39, 0, 1, "", "PytorchExportSerializationFormat"], [39, 0, 1, "", "QuantizationFormat"], [39, 0, 1, "", "keras_export_model"], [39, 0, 1, "", "pytorch_export_model"]], "model_compression_toolkit.gptq": [[4, 0, 1, "", "GPTQHessianScoresConfig"], [4, 0, 1, "", "GradientPTQConfig"], [4, 0, 1, "", "RoundingType"], [15, 3, 1, "", "get_keras_gptq_config"], [17, 3, 1, "", "get_pytorch_gptq_config"], [20, 3, 1, "", "keras_gradient_post_training_quantization"], [28, 3, 1, "", "pytorch_gradient_post_training_quantization"]], "model_compression_toolkit.pruning": [[7, 0, 1, "", "ChannelsFilteringStrategy"], [7, 0, 1, "", "ImportanceMetric"], [7, 0, 1, "", "PruningConfig"], [8, 0, 1, "", "PruningInfo"], [24, 3, 1, "", "keras_pruning_experimental"], [31, 3, 1, "", "pytorch_pruning_experimental"]], "model_compression_toolkit.pruning.PruningConfig": [[7, 2, 1, "", "channels_filtering_strategy"], [7, 2, 1, "", "importance_metric"], [7, 2, 1, "", "num_score_approximations"]], "model_compression_toolkit.pruning.PruningInfo": [[8, 4, 1, "", "importance_scores"], [8, 4, 1, "", "pruning_masks"]], "model_compression_toolkit.ptq": [[23, 3, 1, "", "keras_post_training_quantization"], [30, 3, 1, "", "pytorch_post_training_quantization"]], "model_compression_toolkit.qat": [[42, 0, 1, "", "QATConfig"], [42, 0, 1, "", "TrainingMethod"], [25, 3, 1, "", "keras_quantization_aware_training_finalize_experimental"], [26, 3, 1, "", "keras_quantization_aware_training_init_experimental"], [32, 3, 1, "", "pytorch_quantization_aware_training_finalize_experimental"], [33, 3, 1, "", "pytorch_quantization_aware_training_init_experimental"]], "model_compression_toolkit.target_platform": [[40, 0, 1, "", "AttributeFilter"], [43, 0, 1, "", "AttributeQuantizationConfig"], [40, 0, 1, "", "Eq"], [43, 0, 1, "", "Fusing"], [40, 0, 1, "", "Greater"], [40, 0, 1, "", "GreaterEq"], [43, 0, 1, "", "LayerFilterParams"], [40, 0, 1, "", "NotEq"], [43, 0, 1, "", "OpQuantizationConfig"], [43, 0, 1, "", "OperationsSetToLayers"], [43, 0, 1, "", "OperationsToLayers"], [43, 0, 1, "", "OperatorSetConcat"], [43, 0, 1, "", "OperatorsSet"], [43, 0, 1, "", "QuantizationConfigOptions"], [43, 0, 1, "", "QuantizationMethod"], [40, 0, 1, "", "Smaller"], [40, 0, 1, "", "SmallerEq"], [43, 0, 1, "", "TargetPlatformCapabilities"], [43, 0, 1, "", "TargetPlatformModel"]], "model_compression_toolkit.trainable_infrastructure": [[44, 0, 1, "", "BaseKerasTrainableQuantizer"], [44, 0, 1, "", "BasePytorchTrainableQuantizer"], [44, 0, 1, "", "TrainableQuantizerActivationConfig"], [44, 0, 1, "", "TrainableQuantizerWeightsConfig"]], "model_compression_toolkit.xquant.common.xquant_config": [[12, 0, 1, "", "XQuantConfig"]], "model_compression_toolkit.xquant.keras.facade_xquant_report": [[35, 3, 1, "", "xquant_report_keras_experimental"]], "model_compression_toolkit.xquant.pytorch.facade_xquant_report": [[36, 3, 1, "", "xquant_report_pytorch_experimental"]]}, "objtypes": {"0": "py:class", "1": "py:method", "2": "py:attribute", "3": "py:function", "4": "py:property"}, "objnames": {"0": ["py", "class", "Python class"], "1": ["py", "method", "Python method"], "2": ["py", "attribute", "Python attribute"], "3": ["py", "function", "Python function"], "4": ["py", "property", "Python property"]}, "titleterms": {"bitwidthconfig": 0, "manualbitwidthselect": 0, "data": [1, 19, 27], "gener": [1, 19, 27], "configur": [1, 7, 12, 46], "imagegranular": 1, "schedulertyp": 1, "batchnormalignemntlosstyp": 1, "outputlosstyp": 1, "datainittyp": 1, "bnlayerweightingtyp": 1, "imagepipelinetyp": 1, "imagenormalizationtyp": 1, "defaultdict": 2, "class": [2, 3, 4], "frameworkinfo": 3, "channelaxi": 3, "gradientptqconfig": [4, 15, 17], "gptqhessianscoresconfig": 4, "roundingtyp": 4, "mixedprecisionquantizationconfig": 5, "mpdistanceweight": 6, "prune": [7, 8, 13, 24, 31], "importancemetr": 7, "channelsfilteringstrategi": 7, "inform": [8, 21, 29], "quantizationconfig": 9, "quantizationerrormethod": 10, "resourceutil": 11, "xquant": [12, 13, 35, 36], "api": [13, 47], "doc": 13, "ptq": 13, "gptq": 13, "qat": 13, "core": 13, "data_gener": 13, "export": [13, 39], "trainable_infrastructur": [13, 44], "set_log_fold": 13, "keras_load_quantized_model": 13, "target_platform": [13, 43], "indic": 13, "tabl": 13, "get": [14, 15, 16, 17, 18, 21, 29], "datagenerationconfig": [14, 16], "kera": [14, 15, 19, 20, 21, 22, 23, 24, 25, 26, 35, 39], "model": [14, 15, 16, 17, 21, 22, 25, 26, 29, 32, 33, 39, 47], "pytorch": [16, 17, 27, 28, 29, 30, 31, 32, 33, 36, 39], "targetplatformcap": [18, 43], "gradient": [20, 28], "base": [20, 28], "post": [20, 23, 28, 30], "train": [20, 23, 25, 26, 28, 30, 32, 33], "quantiz": [20, 22, 23, 25, 26, 28, 30, 32, 33, 39], "resourc": [21, 29], "util": [21, 29], "load": 22, "structur": [24, 31], "awar": [25, 26, 32, 33], "final": [25, 32], "init": [26, 33], "enabl": 34, "logger": 34, "report": [35, 36], "coreconfig": 37, "debug_config": 38, "modul": [38, 39, 41, 42, 43, 44], "debugconfig": 38, "quantizationformat": 39, "kerasexportserializationformat": 39, "keras_export_model": 39, "tutori": 39, "serial": 39, "format": 39, "mctq": 39, "pytorchexportserializationformat": 39, "pytorch_export_model": 39, "onnx": 39, "opset": 39, "version": 39, "us": 39, "infer": 39, "layer": 40, "attribut": 40, "filter": [40, 41], "network_editor": 41, "editrul": 41, "action": 41, "qat_config": 42, "trainingmethod": 42, "qatconfig": 42, "quantizationmethod": 43, "opquantizationconfig": 43, "attributequantizationconfig": 43, "quantizationconfigopt": 43, "targetplatformmodel": 43, "operatorsset": 43, "fuse": 43, "operatorsetconcat": 43, "operationstolay": 43, "operationssettolay": 43, "layerfilterparam": 43, "basekerastrainablequant": 44, "basepytorchtrainablequant": 44, "trainablequantizerweightsconfig": 44, "trainablequantizeractivationconfig": 44, "visual": 46, "within": 46, "tensorboard": 46, "cosin": 46, "similar": 46, "comparison": 46, "mix": 46, "precis": 46, "bit": 46, "width": 46, "compress": 47, "toolkit": 47, "user": 47, "guid": 47, "overview": 47, "instal": 47, "support": 47, "featur": 47, "quickstart": 47, "document": 47, "technic": 47, "constraint": 47, "refer": 47}, "envversion": {"sphinx.domains.c": 2, "sphinx.domains.changeset": 1, "sphinx.domains.citation": 1, "sphinx.domains.cpp": 6, "sphinx.domains.index": 1, "sphinx.domains.javascript": 2, "sphinx.domains.math": 2, "sphinx.domains.python": 3, "sphinx.domains.rst": 2, "sphinx.domains.std": 2, "sphinx": 56}}) \ No newline at end of file diff --git a/docs/static/bizstyle.js b/docs/static/bizstyle.js index 344a98362..f41af42bc 100644 --- a/docs/static/bizstyle.js +++ b/docs/static/bizstyle.js @@ -23,7 +23,7 @@ const initialiseBizStyle = () => { } window.addEventListener("resize", - () => (document.querySelector("li.nav-item-0 a").innerText = (window.innerWidth <= 776) ? "Top" : "MCT Documentation: ver 2.1.0") + () => (document.querySelector("li.nav-item-0 a").innerText = (window.innerWidth <= 776) ? "Top" : "MCT Documentation: ver 2.2.0") ) if (document.readyState !== "loading") initialiseBizStyle() diff --git a/docs/static/documentation_options.js b/docs/static/documentation_options.js index 437af4a26..15876331c 100644 --- a/docs/static/documentation_options.js +++ b/docs/static/documentation_options.js @@ -1,6 +1,6 @@ var DOCUMENTATION_OPTIONS = { URL_ROOT: document.getElementById("documentation_options").getAttribute('data-url_root'), - VERSION: '2.1.0', + VERSION: '2.2.0', LANGUAGE: 'en', COLLAPSE_INDEX: false, BUILDER: 'html', diff --git a/docsrc/source/api/api_docs/classes/BitWidthConfig.rst b/docsrc/source/api/api_docs/classes/BitWidthConfig.rst new file mode 100644 index 000000000..9b5cff54c --- /dev/null +++ b/docsrc/source/api/api_docs/classes/BitWidthConfig.rst @@ -0,0 +1,14 @@ +:orphan: + +.. _ug-BitWidthConfig: + + +BitWidthConfig +========================== + +.. autoclass:: model_compression_toolkit.core.BitWidthConfig + +ManualBitWidthSelection +========================== + +.. autoclass:: model_compression_toolkit.core.common.quantization.bit_width_config.ManualBitWidthSelection diff --git a/docsrc/source/api/api_docs/classes/XQuantConfig.rst b/docsrc/source/api/api_docs/classes/XQuantConfig.rst new file mode 100644 index 000000000..79d34128a --- /dev/null +++ b/docsrc/source/api/api_docs/classes/XQuantConfig.rst @@ -0,0 +1,14 @@ +:orphan: + +.. _ug-XQuantConfig: + +================================================ +XQuant Configuration +================================================ + +.. autoclass:: model_compression_toolkit.xquant.common.xquant_config.XQuantConfig + :members: + + + + diff --git a/docsrc/source/api/api_docs/index.rst b/docsrc/source/api/api_docs/index.rst index b349d3be4..0c4433163 100644 --- a/docsrc/source/api/api_docs/index.rst +++ b/docsrc/source/api/api_docs/index.rst @@ -49,6 +49,7 @@ core - :ref:`QuantizationConfig`: Module to configure the quantization process. - :ref:`QuantizationErrorMethod`: Select a method for quantization parameters' selection. - :ref:`MixedPrecisionQuantizationConfig`: Module to configure the quantization process when using mixed-precision PTQ. +- :ref:`BitWidthConfig`: Module to configure the bit-width manually. - :ref:`ResourceUtilization`: Module to configure resources to use when searching for a configuration for the optimized model. - :ref:`MpDistanceWeighting`: Mixed precision distance metric weighting methods. - :ref:`network_editor`: Module to modify the optimization process for troubleshooting. @@ -75,6 +76,13 @@ pruning - :ref:`PruningConfig`: Configuration for the pruning process (experimental). - :ref:`PruningInfo`: Information about the pruned model such as pruned channel indices, etc. (experimental). +xquant +=========== + +- :ref:`xquant_report_pytorch_experimental`: A function to generate an explainable quantization report for a quantized Pytorch model (experimental). +- :ref:`xquant_report_keras_experimental`: A function to generate an explainable quantization report for a quantized Keras model (experimental). + +- :ref:`XQuantConfig`: Configuration for the XQuant report (experimental). exporter ========= diff --git a/docsrc/source/api/api_docs/methods/xquant_report_keras_experimental.rst b/docsrc/source/api/api_docs/methods/xquant_report_keras_experimental.rst new file mode 100644 index 000000000..af6994d1d --- /dev/null +++ b/docsrc/source/api/api_docs/methods/xquant_report_keras_experimental.rst @@ -0,0 +1,12 @@ +:orphan: + +.. _ug-xquant_report_keras_experimental: + + +================================================ +XQuant Report Keras +================================================ + +.. autofunction:: model_compression_toolkit.xquant.keras.facade_xquant_report.xquant_report_keras_experimental + + diff --git a/docsrc/source/api/api_docs/methods/xquant_report_pytorch_experimental.rst b/docsrc/source/api/api_docs/methods/xquant_report_pytorch_experimental.rst new file mode 100644 index 000000000..074db79d9 --- /dev/null +++ b/docsrc/source/api/api_docs/methods/xquant_report_pytorch_experimental.rst @@ -0,0 +1,15 @@ +:orphan: + +.. _ug-xquant_report_pytorch_experimental: + + +================================================ +XQuant Report Pytorch +================================================ + +.. autofunction:: model_compression_toolkit.xquant.pytorch.facade_xquant_report.xquant_report_pytorch_experimental + + + + + diff --git a/model_compression_toolkit/__init__.py b/model_compression_toolkit/__init__.py index 8505d556a..4d45628ef 100644 --- a/model_compression_toolkit/__init__.py +++ b/model_compression_toolkit/__init__.py @@ -27,4 +27,4 @@ from model_compression_toolkit import pruning from model_compression_toolkit.trainable_infrastructure.keras.load_model import keras_load_quantized_model -__version__ = "2.1.0" +__version__ = "2.2.0" diff --git a/model_compression_toolkit/xquant/common/xquant_config.py b/model_compression_toolkit/xquant/common/xquant_config.py index 94c35b498..624822812 100644 --- a/model_compression_toolkit/xquant/common/xquant_config.py +++ b/model_compression_toolkit/xquant/common/xquant_config.py @@ -31,9 +31,7 @@ def __init__(self, Args: report_dir (str): Directory where the reports will be saved. - custom_similarity_metrics (Dict[str, Callable]): Custom similarity metrics to be computed between tensors - of the two models. The dictionary keys are similarity metric names and the values are callables that implement the - similarity metric computation. + custom_similarity_metrics (Dict[str, Callable]): Custom similarity metrics to be computed between tensors of the two models. The dictionary keys are similarity metric names and the values are callables that implement the similarity metric computation. """ self.report_dir = report_dir self.custom_similarity_metrics = custom_similarity_metrics