da4ml.converter.hgq2 package

Submodules

da4ml.converter.hgq2.parser module

class da4ml.converter.hgq2.parser.OpObj(operation: keras.src.ops.operation.Operation, args: list, kwargs: dict, produces: tuple[keras.src.backend.common.keras_tensor.KerasTensor, ...], requires: tuple[keras.src.backend.common.keras_tensor.KerasTensor, ...])

Bases: object

args: list

kwargs: dict

operation: Operation

produces: tuple[KerasTensor, ...]

requires: tuple[KerasTensor, ...]

da4ml.converter.hgq2.parser.parse_model(model: Model)

da4ml.converter.hgq2.parser.replace_tensors(tensor_map: dict[KerasTensor, FixedVariableArray], obj: Any) → Any

da4ml.converter.hgq2.parser.trace_model(model: Model, hwconf: HWConfig = HWConfig(1, -1, -1), solver_options: dict[str, Any] | None = None, verbose: bool = False, inputs: tuple[FixedVariableArray, ...] | FixedVariableArray | None = None, dump: Literal[False] = False) → tuple[FixedVariableArray, FixedVariableArray]
da4ml.converter.hgq2.parser.trace_model(model: Model, hwconf: HWConfig = HWConfig(1, -1, -1), solver_options: dict[str, Any] | None = None, verbose: bool = False, inputs: tuple[FixedVariableArray, ...] | FixedVariableArray | None = None, dump: Literal[True] = False) → dict[str, FixedVariableArray]

da4ml.converter.hgq2.replica module

class da4ml.converter.hgq2.replica.ReplayAbs(layer: Operation)

Bases: ReplayOperationBase

call(x: FixedVariableArray) → FixedVariableArray

handles: tuple[type, ...] = (<class 'keras.src.ops.numpy.Absolute'>, <class 'keras.src.ops.numpy.Abs'>)

class da4ml.converter.hgq2.replica.ReplayArithmetic(layer: Operation)

Bases: ReplayOperationBase

call(x1: FixedVariableArray, x2: FixedVariableArray)

handles: tuple[type, ...] = (<class 'keras.src.ops.numpy.Add'>, <class 'keras.src.ops.numpy.Subtract'>, <class 'keras.src.ops.numpy.Multiply'>, <class 'keras.src.ops.numpy.TrueDivide'>, <class 'keras.src.ops.numpy.Divide'>, <class 'hgq.layers.ops.merge.QSubtract'>, <class 'hgq.layers.ops.merge.QMaximum'>, <class 'hgq.layers.ops.merge.QMinimum'>, <class 'keras.src.ops.numpy.Maximum'>, <class 'keras.src.ops.numpy.Minimum'>)

class da4ml.converter.hgq2.replica.ReplayConcatenate(layer: Operation)

Bases: ReplayOperationBase

call(xs: Sequence[FixedVariableArray])

handles: tuple[type, ...] = (<class 'keras.src.ops.numpy.Concatenate'>,)

class da4ml.converter.hgq2.replica.ReplayEinsum(layer: Operation)

Bases: ReplayOperationBase

call(*operands: FixedVariableArray) → FixedVariableArray

handles: tuple[type, ...] = (<class 'keras.src.ops.numpy.Einsum'>,)

class da4ml.converter.hgq2.replica.ReplayGetItem(layer: Operation)

Bases: ReplayOperationBase

call(x: FixedVariableArray, key)

handles: tuple[type, ...] = (<class 'keras.src.ops.numpy.GetItem'>,)

class da4ml.converter.hgq2.replica.ReplayMatmul(layer: Operation)

Bases: ReplayOperationBase

call(x1: FixedVariableArray, x2: FixedVariableArray) → FixedVariableArray

handles: tuple[type, ...] = (<class 'keras.src.ops.numpy.Matmul'>, <class 'keras.src.ops.numpy.Dot'>)

class da4ml.converter.hgq2.replica.ReplayMerge(layer: Operation)

Bases: ReplayOperationBase

call(inputs: tuple[FixedVariableArray, FixedVariableArray]) → FixedVariableArray

handles: tuple[type, ...] = (<class 'keras.src.layers.merging.add.Add'>, <class 'keras.src.layers.merging.concatenate.Concatenate'>, <class 'hgq.layers.ops.merge.QAdd'>)

class da4ml.converter.hgq2.replica.ReplayMoveaxis(layer: Operation)

Bases: ReplayOperationBase

call(x: FixedVariableArray)

handles: tuple[type, ...] = (<class 'keras.src.ops.numpy.Moveaxis'>,)

class da4ml.converter.hgq2.replica.ReplayNoOp(layer: Operation)

Bases: ReplayOperationBase

call(x: FixedVariableArray, training=False) → FixedVariableArray

handles: tuple[type, ...] = (<class 'keras.src.layers.preprocessing.image_preprocessing.random_brightness.RandomBrightness'>, <class 'keras.src.layers.preprocessing.image_preprocessing.random_color_degeneration.RandomColorDegeneration'>, <class 'keras.src.layers.preprocessing.image_preprocessing.random_color_jitter.RandomColorJitter'>, <class 'keras.src.layers.preprocessing.image_preprocessing.random_contrast.RandomContrast'>, <class 'keras.src.layers.preprocessing.image_preprocessing.random_crop.RandomCrop'>, <class 'keras.src.layers.preprocessing.image_preprocessing.random_elastic_transform.RandomElasticTransform'>, <class 'keras.src.layers.preprocessing.image_preprocessing.random_erasing.RandomErasing'>, <class 'keras.src.layers.preprocessing.image_preprocessing.random_flip.RandomFlip'>, <class 'keras.src.layers.preprocessing.image_preprocessing.random_gaussian_blur.RandomGaussianBlur'>, <class 'keras.src.layers.preprocessing.image_preprocessing.random_grayscale.RandomGrayscale'>, <class 'keras.src.layers.preprocessing.image_preprocessing.random_hue.RandomHue'>, <class 'keras.src.layers.preprocessing.image_preprocessing.random_invert.RandomInvert'>, <class 'keras.src.layers.preprocessing.image_preprocessing.random_perspective.RandomPerspective'>, <class 'keras.src.layers.preprocessing.image_preprocessing.random_posterization.RandomPosterization'>, <class 'keras.src.layers.preprocessing.image_preprocessing.random_rotation.RandomRotation'>, <class 'keras.src.layers.preprocessing.image_preprocessing.random_saturation.RandomSaturation'>, <class 'keras.src.layers.preprocessing.image_preprocessing.random_sharpness.RandomSharpness'>, <class 'keras.src.layers.preprocessing.image_preprocessing.random_shear.RandomShear'>, <class 'keras.src.layers.preprocessing.image_preprocessing.random_translation.RandomTranslation'>, <class 'keras.src.layers.preprocessing.image_preprocessing.random_zoom.RandomZoom'>, <class 'keras.src.layers.regularization.alpha_dropout.AlphaDropout'>, <class 'keras.src.layers.regularization.dropout.Dropout'>, <class 'keras.src.layers.regularization.gaussian_dropout.GaussianDropout'>, <class 'keras.src.layers.regularization.gaussian_noise.GaussianNoise'>, <class 'keras.src.layers.regularization.spatial_dropout.SpatialDropout1D'>, <class 'keras.src.layers.regularization.spatial_dropout.SpatialDropout2D'>, <class 'keras.src.layers.regularization.spatial_dropout.SpatialDropout3D'>)

class da4ml.converter.hgq2.replica.ReplayOperationBase(layer: Operation)

Bases: object

call(*args, **kwargs) → tuple[FixedVariableArray, ...] | FixedVariableArray

handles: tuple[type, ...] = ()

class da4ml.converter.hgq2.replica.ReplayOperationMeta(name: str, bases: tuple[type, ...], namespace: dict[str, Any]): Bases: type

class da4ml.converter.hgq2.replica.ReplayPool(layer: Operation)

Bases: ReplayOperationBase

call(inputs: FixedVariableArray) → FixedVariableArray

handles: tuple[type, ...] = (<class 'hgq.layers.pooling.QAveragePooling1D'>, <class 'hgq.layers.pooling.QAveragePooling2D'>, <class 'hgq.layers.pooling.QAveragePooling3D'>, <class 'hgq.layers.pooling.QMaxPooling1D'>, <class 'hgq.layers.pooling.QMaxPooling2D'>, <class 'hgq.layers.pooling.QMaxPooling3D'>, <class 'hgq.layers.pooling.QGlobalAveragePooling1D'>, <class 'hgq.layers.pooling.QGlobalMaxPooling1D'>, <class 'hgq.layers.pooling.QGlobalAveragePooling2D'>, <class 'hgq.layers.pooling.QGlobalMaxPooling2D'>, <class 'hgq.layers.pooling.QGlobalAveragePooling3D'>, <class 'hgq.layers.pooling.QGlobalMaxPooling3D'>, <class 'keras.src.layers.pooling.average_pooling1d.AveragePooling1D'>, <class 'keras.src.layers.pooling.average_pooling2d.AveragePooling2D'>, <class 'keras.src.layers.pooling.average_pooling3d.AveragePooling3D'>, <class 'keras.src.layers.pooling.max_pooling1d.MaxPooling1D'>, <class 'keras.src.layers.pooling.max_pooling2d.MaxPooling2D'>, <class 'keras.src.layers.pooling.max_pooling3d.MaxPooling3D'>, <class 'keras.src.layers.pooling.global_average_pooling1d.GlobalAveragePooling1D'>, <class 'keras.src.layers.pooling.global_max_pooling1d.GlobalMaxPooling1D'>, <class 'keras.src.layers.pooling.global_average_pooling2d.GlobalAveragePooling2D'>, <class 'keras.src.layers.pooling.global_max_pooling2d.GlobalMaxPooling2D'>, <class 'keras.src.layers.pooling.global_average_pooling3d.GlobalAveragePooling3D'>, <class 'keras.src.layers.pooling.global_max_pooling3d.GlobalMaxPooling3D'>)

op: Any

class da4ml.converter.hgq2.replica.ReplayQBatchNormalization(layer: Operation)

Bases: ReplayOperationBase

call(inputs: FixedVariableArray) → FixedVariableArray

handles: tuple[type, ...] = (<class 'hgq.layers.batch_normalization.QBatchNormalization'>,)

op: Any

class da4ml.converter.hgq2.replica.ReplayQConv(layer: Operation)

Bases: ReplayOperationBase

call(inputs: FixedVariableArray) → FixedVariableArray

handles: tuple[type, ...] = (<class 'hgq.layers.conv.QConv1D'>, <class 'hgq.layers.conv.QConv2D'>, <class 'hgq.layers.conv.QConv3D'>)

op: Any

class da4ml.converter.hgq2.replica.ReplayQDense(layer: Operation)

Bases: ReplayOperationBase

call(inputs: FixedVariableArray) → FixedVariableArray

handles: tuple[type, ...] = (<class 'hgq.layers.core.dense.QDense'>, <class 'hgq.layers.core.einsum_dense.QEinsumDense'>, <class 'hgq.layers.einsum_dense_batchnorm.QEinsumDenseBatchnorm'>, <class 'hgq.layers.core.dense.QBatchNormDense'>, <class 'keras.src.layers.core.einsum_dense.EinsumDense'>)

op: Any

class da4ml.converter.hgq2.replica.ReplayQDot(layer: Operation)

Bases: ReplayOperationBase

call(inputs: tuple[FixedVariableArray, FixedVariableArray]) → FixedVariableArray

handles: tuple[type, ...] = (<class 'hgq.layers.ops.merge.QDot'>, <class 'keras.src.layers.merging.dot.Dot'>)

op: Any

class da4ml.converter.hgq2.replica.ReplayQEinsum(layer: Operation)

Bases: ReplayOperationBase

call(inputs: tuple[FixedVariableArray, ...]) → FixedVariableArray

handles: tuple[type, ...] = (<class 'hgq.layers.ops.einsum.QEinsum'>,)

op: Any

class da4ml.converter.hgq2.replica.ReplayQReduction(layer: Operation)

Bases: ReplayOperationBase

call(x: FixedVariableArray)

handles: tuple[type, ...] = (<class 'hgq.layers.ops.accum.QSum'>, <class 'hgq.layers.ops.accum.QMeanPow2'>)

op: Any

class da4ml.converter.hgq2.replica.ReplayQuantizer(op: Quantizer)

Bases: ReplayOperationBase

call(inputs: FixedVariableArray) → FixedVariableArray

handles: tuple[type, ...] = (<class 'hgq.quantizer.quantizer.Quantizer'>,)

op: Any

class da4ml.converter.hgq2.replica.ReplayReLU(layer: Operation)

Bases: ReplayOperationBase

call(inputs: FixedVariableArray) → FixedVariableArray

handles: tuple[type, ...] = (<class 'keras.src.layers.activations.relu.ReLU'>,)

op: Any

class da4ml.converter.hgq2.replica.ReplayReduction(layer: Operation)

Bases: ReplayOperationBase

call(x: FixedVariableArray, axis=None, keepdims=False)

handles: tuple[type, ...] = (<class 'keras.src.ops.numpy.Sum'>, <class 'keras.src.ops.numpy.Max'>, <class 'keras.src.ops.numpy.Min'>)

op: Any

class da4ml.converter.hgq2.replica.ReplayRepeat(layer: Operation)

Bases: ReplayOperationBase

call(x: FixedVariableArray)

handles: tuple[type, ...] = (<class 'keras.src.ops.numpy.Repeat'>,)

op: Any

class da4ml.converter.hgq2.replica.ReplayRepeatVector(layer: Operation)

Bases: ReplayOperationBase

call(inputs: FixedVariableArray) → FixedVariableArray

handles: tuple[type, ...] = (<class 'keras.src.layers.reshaping.repeat_vector.RepeatVector'>,)

op: Any

class da4ml.converter.hgq2.replica.ReplayReshape(layer: Operation)

Bases: ReplayOperationBase

call(inputs: FixedVariableArray) → FixedVariableArray

handles: tuple[type, ...] = (<class 'keras.src.layers.reshaping.reshape.Reshape'>, <class 'keras.src.layers.reshaping.flatten.Flatten'>, <class 'keras.src.ops.numpy.Reshape'>, <class 'keras.src.ops.numpy.Ravel'>)

op: Any

class da4ml.converter.hgq2.replica.ReplayTranspose(layer: Operation)

Bases: ReplayOperationBase

call(x: FixedVariableArray)

handles: tuple[type, ...] = (<class 'keras.src.ops.numpy.Transpose'>,)

op: Any

da4ml.converter.hgq2.replica.mirror_quantizer(q: Quantizer, v: FixedVariableArray) → FixedVariableArray

Module contents

da4ml.converter.hgq2.trace_model(model: Model, hwconf: HWConfig = (1, -1, -1), solver_options: dict[str, Any] | None = None, verbose: bool = False, inputs: tuple[FixedVariableArray, ...] | None = None, dump=False)