diff options
Diffstat (limited to 'src/mlia/nn')
-rw-r--r-- | src/mlia/nn/rewrite/core/rewrite.py | 251 | ||||
-rw-r--r-- | src/mlia/nn/rewrite/core/train.py | 120 | ||||
-rw-r--r-- | src/mlia/nn/rewrite/library/clustering.py | 79 | ||||
-rw-r--r-- | src/mlia/nn/rewrite/library/fc_clustering_layer.py | 26 | ||||
-rw-r--r-- | src/mlia/nn/rewrite/library/fc_layer.py | 18 | ||||
-rw-r--r-- | src/mlia/nn/rewrite/library/fc_sparsity24_layer.py | 23 | ||||
-rw-r--r-- | src/mlia/nn/rewrite/library/helper_functions.py | 61 | ||||
-rw-r--r-- | src/mlia/nn/rewrite/library/layers.py | 53 | ||||
-rw-r--r-- | src/mlia/nn/rewrite/library/sparsity.py | 146 | ||||
-rw-r--r-- | src/mlia/nn/select.py | 24 |
10 files changed, 653 insertions, 148 deletions
diff --git a/src/mlia/nn/rewrite/core/rewrite.py b/src/mlia/nn/rewrite/core/rewrite.py index e2c097c..a6bd306 100644 --- a/src/mlia/nn/rewrite/core/rewrite.py +++ b/src/mlia/nn/rewrite/core/rewrite.py @@ -8,13 +8,20 @@ import tempfile from abc import ABC from abc import abstractmethod from dataclasses import dataclass +from inspect import getfullargspec from pathlib import Path +from statistics import fmean from typing import Any from typing import Callable +from typing import Generator import numpy as np +import tensorflow as tf import tensorflow_model_optimization as tfmot from keras.api._v2 import keras # Temporary workaround for now: MLIA-1107 +from tensorflow_model_optimization.python.core.sparsity.keras.pruning_utils import ( # pylint: disable=no-name-in-module + is_pruned_m_by_n, +) from mlia.core.errors import ConfigurationError from mlia.core.reporting import Column @@ -24,33 +31,57 @@ from mlia.nn.common import Optimizer from mlia.nn.common import OptimizerConfiguration from mlia.nn.rewrite.core.train import train from mlia.nn.rewrite.core.train import TrainingParameters -from mlia.nn.rewrite.library.fc_clustering_layer import ( - get_keras_model_clus as fc_clustering_rewrite, -) -from mlia.nn.rewrite.library.fc_layer import get_keras_model as fc_rewrite -from mlia.nn.rewrite.library.fc_sparsity24_layer import ( - get_keras_model as fc_rewrite_sparsity24, -) +from mlia.nn.rewrite.library.clustering import conv2d_clustering_rewrite +from mlia.nn.rewrite.library.clustering import fc_clustering_rewrite +from mlia.nn.rewrite.library.layers import conv2d_rewrite +from mlia.nn.rewrite.library.layers import fc_rewrite +from mlia.nn.rewrite.library.sparsity import conv2d_sparsity_rewrite +from mlia.nn.rewrite.library.sparsity import conv2d_sparsity_unstructured_rewrite +from mlia.nn.rewrite.library.sparsity import fc_sparsity_rewrite +from mlia.nn.rewrite.library.sparsity import fc_sparsity_unstructured_rewrite from mlia.nn.tensorflow.config import TFLiteModel from mlia.utils.registry import Registry - logger = logging.getLogger(__name__) -RewriteCallable = Callable[[Any, Any], keras.Model] +RewriteCallable = Callable[..., keras.Model] class Rewrite(ABC): """Abstract class for rewrite logic to be used by RewritingOptimizer.""" - def __init__(self, name: str, rewrite_fn: RewriteCallable): + def __init__( + self, + name: str, + rewrite_fn: RewriteCallable, + rewrite_fn_extra_args: dict[str, Any] | None = None, + ): """Initialize a Rewrite instance with a given name and an optional function.""" self.name = name self.function = rewrite_fn + self.function_extra_args = rewrite_fn_extra_args - def __call__(self, input_shape: Any, output_shape: Any) -> keras.Model: + def __call__( + self, input_shape: Any, output_shape: Any, **kwargs: Any + ) -> keras.Model: """Perform the rewrite operation using the configured function.""" try: - return self.function(input_shape, output_shape) + if self.function_extra_args: + return self.function( + input_shape, output_shape, **kwargs, **self.function_extra_args + ) + + return self.function(input_shape, output_shape, **kwargs) + except TypeError as ex: + expected_args = self.return_rewrite_func_args() + if "input_shape" in expected_args: + expected_args.remove("input_shape") + if "output_shape" in expected_args: + expected_args.remove("output_shape") + raise KeyError( + f"Found unexpected parameters for rewrite. Expected (sub)set " + f"of {expected_args} found unexpected parameter(s) " + f"{list(set(list(kwargs.keys())) - set(expected_args))}" + ) from ex except Exception as ex: raise RuntimeError(f"Rewrite '{self.name}' failed.") from ex @@ -58,21 +89,25 @@ class Rewrite(ABC): """Return a quantized model if required.""" return model + def return_rewrite_func_args(self) -> list[str]: + """Return the expected args of the rewrite function.""" + return getfullargspec(self.function).args + @abstractmethod def training_callbacks(self) -> list: - """Return default rewrite callbacks.""" + """Return rewrite callbacks.""" @abstractmethod def post_process(self, model: keras.Model) -> keras.Model: - """Return default post-processing rewrite options.""" + """Return post-processing rewrite option.""" @abstractmethod - def check_optimization(self, model: keras.Model, **kwargs: dict) -> bool: + def check_optimization(self, model: keras.Model) -> bool: """Check if the optimization has produced the correct result.""" class GenericRewrite(Rewrite): - """Graph rewrite logic for fully-connected rewrite.""" + """Rewrite class for generic rewrites e.g. fully-connected.""" def quantize(self, model: keras.Model) -> keras.Model: """Return a quantized model if required.""" @@ -83,10 +118,10 @@ class GenericRewrite(Rewrite): return [] def post_process(self, model: keras.Model) -> keras.Model: - """Return default post-processing rewrite options.""" + """Return default post-processing rewrite option.""" return model - def check_optimization(self, model: keras.Model, **kwargs: Any) -> bool: + def check_optimization(self, model: keras.Model, **_: Any) -> bool: """Not needed here.""" return True @@ -99,16 +134,31 @@ class QuantizeAwareTrainingRewrite(Rewrite, ABC): """Apply optimization-aware quantization to a given model.""" return model + def check_optimization_generator( + self, model: keras.Model + ) -> Generator[tuple[tf.Tensor, keras.layers.Layer], None, None]: + """Loop for check_optimization function.""" + for layer in model.layers: + for weight in layer.weights: + if "kernel" in weight.name: + if ( + "kernel_min" in weight.name + or "kernel_max" in weight.name + or "depthwise" in weight.name + ): + continue + yield weight, layer + -class Sparsity24Rewrite(QuantizeAwareTrainingRewrite): - """Graph rewrite logic for fully-connected-sparsity24 rewrite.""" +class SparsityRewrite(QuantizeAwareTrainingRewrite): + """Base rewrite class for sparsity rewrites.""" pruning_callback = tfmot.sparsity.keras.UpdatePruningStep strip_pruning_wrapper = staticmethod(tfmot.sparsity.keras.strip_pruning) def quantize(self, model: keras.Model) -> keras.Model: - """Skip quantization when using pruning rewrite.""" + """Skip quantization when using sparsity rewrite.""" return model def training_callbacks(self) -> list: @@ -116,7 +166,7 @@ class Sparsity24Rewrite(QuantizeAwareTrainingRewrite): return [self.pruning_callback()] def post_process(self, model: keras.Model) -> keras.Model: - """Pruning-specific post-processing rewrite options.""" + """Pruning-specific post-processing rewrite option.""" return self.strip_pruning_wrapper(model) def preserved_quantize( @@ -129,16 +179,78 @@ class Sparsity24Rewrite(QuantizeAwareTrainingRewrite): model, tfmot.experimental.combine.Default8BitPrunePreserveQuantizeScheme(), ) - return model - def check_optimization(self, model: keras.Model, **kwargs: Any) -> bool: + def check_optimization(self, model: keras.Model) -> bool: + """Not needed here.""" + return True + + +class UnstructuredSparsityRewrite(SparsityRewrite): + """ + Rewrite class for unstructured sparsity rewrite. + + e.g. fully-connected-unstructured-sparsity. + """ + + def check_optimization( + self, model: keras.Model, final_sparsity: float = 0.5, **_: Any + ) -> bool: """Not needed here.""" + found_sparsity_list = [] + num_dec_places = str(final_sparsity)[::-1].find(".") + for weight, _ in self.check_optimization_generator(model=model): + weight_np = weight.numpy() + found_sparsity_list.append( + round(np.count_nonzero(weight_np) / weight_np.size, num_dec_places) + ) + if len(found_sparsity_list) == 0: + logger.warning( + "\nWARNING: Could not find any layers " + "in rewrite that could be sparsely pruned" + ) + return False + found_sparsity = fmean(found_sparsity_list) + if found_sparsity != final_sparsity: + logger.warning( + "\nWARNING: Found total sparsity of " + "rewrite model: %.2f " + "expected total sparsity to be: " + "%.2f\n", + found_sparsity, + final_sparsity, + ) + return False + return True + + +class StructuredSparsityRewrite(SparsityRewrite): + """Rewrite class for structured sparsity rewrite e.g. fully-connected-sparsity.""" + + def check_optimization( + self, + model: keras.Model, + sparsity_m: int = 2, + sparsity_n: int = 4, + **_: Any, + ) -> bool: + """Check if sparity has produced the correct result.""" + for weight, layer in self.check_optimization_generator(model=model): + if not is_pruned_m_by_n(weight, m_by_n=(sparsity_m, sparsity_n)): + logger.warning( + "\nWARNING: Could not find (%d, %d) sparsity, " + "in layer %s for weight %s \n", + sparsity_m, + sparsity_n, + layer.name, + weight.name, + ) + return False return True class ClusteringRewrite(QuantizeAwareTrainingRewrite): - """Graph clustering rewrite logic to be used by RewritingOptimizer.""" + """Rewrite class for clustering rewrite e.g. fully-connected-clustering.""" _strip_clustering_wrapper = staticmethod(tfmot.clustering.keras.strip_clustering) @@ -151,32 +263,22 @@ class ClusteringRewrite(QuantizeAwareTrainingRewrite): ) return cqat_model - def check_optimization(self, model: keras.Model, **kwargs: Any) -> bool: + def check_optimization( + self, model: keras.Model, num_clusters: int = 2, **_: Any + ) -> bool: """Check if clustering has produced the correct result.""" - number_of_clusters = kwargs.get("number_of_clusters") - if not number_of_clusters: - raise ValueError( - """ - Expected check_preserved_quantize to have argument number_of_clusters. - """ - ) - - for layer in model.layers: - for weight in layer.weights: - if "kernel" in weight.name: - if "kernel_min" in weight.name or "kernel_max" in weight.name: - continue - number_of_found_clusters = len(np.unique(weight)) - if number_of_found_clusters != number_of_clusters: - logger.warning( - "\nWARNING: Expected %d cluster(s), found %d " - "cluster(s) in layer %s for weight %s \n", - number_of_clusters, - number_of_found_clusters, - layer.name, - weight.name, - ) - return False + for weight, layer in self.check_optimization_generator(model=model): + number_of_found_clusters = len(np.unique(weight)) + if number_of_found_clusters != num_clusters: + logger.warning( + "\nWARNING: Expected %d cluster(s), found %d " + "cluster(s) in layer %s for weight %s \n", + num_clusters, + number_of_found_clusters, + layer.name, + weight.name, + ) + return False return True def training_callbacks(self) -> list: @@ -184,7 +286,7 @@ class ClusteringRewrite(QuantizeAwareTrainingRewrite): return [] def post_process(self, model: keras.Model) -> keras.Model: - """Return the clustering stripped model.""" + """Clustering-specific post-processing rewrite option.""" return self._strip_clustering_wrapper(model) @@ -215,6 +317,7 @@ class RewriteConfiguration(OptimizerConfiguration): layers_to_optimize: list[str] | None = None dataset: Path | None = None train_params: TrainingParameters = TrainingParameters() + rewrite_specific_params: dict | None = None def __str__(self) -> str: """Return string representation of the configuration.""" @@ -227,8 +330,38 @@ class RewritingOptimizer(Optimizer): registry = RewriteRegistry( [ GenericRewrite("fully-connected", fc_rewrite), - Sparsity24Rewrite("fully-connected-sparsity24", fc_rewrite_sparsity24), + GenericRewrite("conv2d", conv2d_rewrite), + GenericRewrite( + "depthwise-separable-conv2d", + conv2d_rewrite, + {"layer_type": keras.layers.SeparableConv2D}, + ), + StructuredSparsityRewrite("fully-connected-sparsity", fc_sparsity_rewrite), ClusteringRewrite("fully-connected-clustering", fc_clustering_rewrite), + ClusteringRewrite("conv2d-clustering", conv2d_clustering_rewrite), + StructuredSparsityRewrite("conv2d-sparsity", conv2d_sparsity_rewrite), + UnstructuredSparsityRewrite( + "conv2d-unstructured-sparsity", conv2d_sparsity_unstructured_rewrite + ), + UnstructuredSparsityRewrite( + "fully-connected-unstructured-sparsity", + fc_sparsity_unstructured_rewrite, + ), + ClusteringRewrite( + "depthwise-separable-conv2d-clustering", + conv2d_clustering_rewrite, + {"layer_type": keras.layers.SeparableConv2D}, + ), + StructuredSparsityRewrite( + "depthwise-separable-conv2d-sparsity", + conv2d_sparsity_rewrite, + {"layer_type": keras.layers.SeparableConv2D}, + ), + UnstructuredSparsityRewrite( + "depthwise-separable-conv2d-unstructured-sparsity", + conv2d_sparsity_unstructured_rewrite, + {"layer_type": keras.layers.SeparableConv2D}, + ), ] ) @@ -250,11 +383,13 @@ class RewritingOptimizer(Optimizer): rewrite = RewritingOptimizer.registry.items[ self.optimizer_configuration.optimization_target ] - use_unmodified_model = True tflite_model = self.model.model_path - tfrecord = str(self.optimizer_configuration.dataset) - + tfrecord = ( + str(self.optimizer_configuration.dataset) + if self.optimizer_configuration.dataset + else None + ) tmp_dir = tempfile.mkdtemp() tmp_output = Path(tmp_dir, "output.tflite") @@ -266,12 +401,16 @@ class RewritingOptimizer(Optimizer): source_model=tflite_model, unmodified_model=tflite_model if use_unmodified_model else None, output_model=str(tmp_output), - input_tfrec=str(tfrecord), + input_tfrec=tfrecord, rewrite=rewrite, is_qat=isinstance(rewrite, QuantizeAwareTrainingRewrite), input_tensors=[self.optimizer_configuration.layers_to_optimize[0]], output_tensors=[self.optimizer_configuration.layers_to_optimize[1]], train_params=self.optimizer_configuration.train_params, + rewrite_specific_params=self.optimizer_configuration.rewrite_specific_params, # pylint: disable=line-too-long + detect_activation_function=( + "activation" in rewrite.return_rewrite_func_args() + ), ) if orig_vs_repl_stats: diff --git a/src/mlia/nn/rewrite/core/train.py b/src/mlia/nn/rewrite/core/train.py index 4204978..b20430b 100644 --- a/src/mlia/nn/rewrite/core/train.py +++ b/src/mlia/nn/rewrite/core/train.py @@ -33,14 +33,15 @@ from mlia.nn.rewrite.core.graph_edit.join import join_models from mlia.nn.rewrite.core.graph_edit.record import record_model from mlia.nn.rewrite.core.utils.numpy_tfrecord import numpytf_count from mlia.nn.rewrite.core.utils.numpy_tfrecord import numpytf_read +from mlia.nn.rewrite.core.utils.numpy_tfrecord import NumpyTFWriter from mlia.nn.rewrite.core.utils.parallel import ParallelTFLiteModel +from mlia.nn.rewrite.library.helper_functions import ACTIVATION_FUNCTION_LIST from mlia.nn.tensorflow.config import TFLiteModel from mlia.nn.tensorflow.tflite_convert import convert_to_tflite from mlia.nn.tensorflow.tflite_graph import load_fb from mlia.nn.tensorflow.tflite_graph import save_fb from mlia.utils.logging import log_action - os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3" tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR) logger = logging.getLogger(__name__) @@ -73,18 +74,48 @@ class TrainingParameters: checkpoint_at: list | None = None +def generate_random_dataset(source_model: str, dataset_path: str) -> str: + """Generate random dataset for model.""" + model = TFLiteModel(model_path=source_model) + input_name = model.input_tensors()[0] + model_is_quantized = model.is_tensor_quantized(name=input_name) + input_shape = model.shape_from_name[input_name][1:] + rand_data_path = dataset_path + "/rand_data.tfrec" + with NumpyTFWriter(rand_data_path) as writer: + for _ in range(5000): + input_data = np.random.rand(1, *input_shape) + input_data = ( + input_data.astype(np.int8) + if model_is_quantized + else input_data.astype(np.float32) + ) + writer.write({input_name: input_data}) + return rand_data_path + + def train( # pylint: disable=too-many-arguments source_model: str, unmodified_model: Any, output_model: str, - input_tfrec: str, + input_tfrec: str | None, rewrite: Callable, is_qat: bool, input_tensors: list, output_tensors: list, train_params: TrainingParameters = TrainingParameters(), + rewrite_specific_params: dict | None = None, + detect_activation_function: bool = False, ) -> Any: """Extract and train a model, and return the results.""" + rand_data_dir_path = None + if not input_tfrec: + logger.info( + "INFO: No dataset given, using random data to perform the rewrite! " + ) + rand_data_dir_path = ( + tempfile.TemporaryDirectory() # pylint: disable=consider-using-with + ) + input_tfrec = generate_random_dataset(source_model, rand_data_dir_path.name) if unmodified_model: unmodified_model_dir = ( tempfile.TemporaryDirectory() # pylint: disable=consider-using-with @@ -122,6 +153,8 @@ def train( # pylint: disable=too-many-arguments rewrite=rewrite, is_qat=is_qat, train_params=train_params, + rewrite_specific_params=rewrite_specific_params, + detect_activation_function=detect_activation_function, ) for i, filename in enumerate(tflite_filenames): @@ -163,7 +196,8 @@ def train( # pylint: disable=too-many-arguments if unmodified_model_dir: cast(tempfile.TemporaryDirectory, unmodified_model_dir).cleanup() - + if rand_data_dir_path: + cast(tempfile.TemporaryDirectory, rand_data_dir_path).cleanup() return results, [ mae, nrmse, @@ -349,6 +383,41 @@ def set_up_data_pipeline( return dataset, steps_per_epoch +def detect_activation_from_rewrite_function(model_path: str) -> str: + """Given a rewrite model, choose the most common activation function.""" + interpreter = tf.lite.Interpreter(model_path=model_path) + interpreter.allocate_tensors() + act_func_match_list = [] + for tensor_details in interpreter.get_tensor_details(): + for act_func in ACTIVATION_FUNCTION_LIST: + tensor_name = tensor_details["name"].lower() + if act_func in tensor_name: + act_func_idx = tensor_name.index(act_func) + if ( + len(tensor_name) == act_func_idx + len(act_func) + or tensor_name[act_func_idx + len(act_func)] == ";" + ): + act_func_match_list.append( + tensor_name[ + act_func_idx : act_func_idx + len(act_func) # noqa: E203 + ] + ) + act_func_match = "relu" + if len(act_func_match_list) == 0: + logger.info( + "No activation function specified, setting activation function to ReLU" + ) + else: + act_func_match = max(set(act_func_match_list), key=act_func_match.count) + logger.info( + "No activation function specified, " + "setting activation function to most " + "common activation detected in rewrite graph: %s", + act_func_match, + ) + return act_func_match + + def train_in_dir( train_dir: str, baseline_dir: Any, @@ -356,6 +425,8 @@ def train_in_dir( rewrite: Callable, is_qat: bool, train_params: TrainingParameters = TrainingParameters(), + rewrite_specific_params: dict | None = None, + detect_activation_function: bool = False, ) -> list[str]: """Train a replacement for replace.tflite using the input.tfrec \ and output.tfrec in train_dir. @@ -372,6 +443,18 @@ def train_in_dir( ) replace = TFLiteModel(ExtractPaths.tflite.replace(train_dir)) + if detect_activation_function and ( + rewrite_specific_params is None + or "activation" not in list(rewrite_specific_params.keys()) + ): + detected_activation_function = detect_activation_from_rewrite_function( + ExtractPaths.tflite.replace(train_dir).as_posix() + ) + if rewrite_specific_params: + rewrite_specific_params["activation"] = detected_activation_function + else: + rewrite_specific_params = {"activation": detected_activation_function} + input_name, output_name = _get_io_tensors(teacher) model_is_quantized = replace.is_tensor_quantized(name=input_name) @@ -396,7 +479,13 @@ def train_in_dir( loss_fn = keras.losses.MeanSquaredError() model = create_model( - rewrite, input_shape, output_shape, optimizer, loss_fn, model_is_quantized + rewrite, + input_shape, + output_shape, + optimizer, + loss_fn, + model_is_quantized, + rewrite_specific_params=rewrite_specific_params, ) logger.info(model.summary()) @@ -462,11 +551,9 @@ def train_in_dir( steps_per_epoch, post_process=True, ) - - # Placeholder for now, will be parametrized later (MLIA-1114) - # rewrite.check_optimization( # type: ignore[attr-defined] - # model, number_of_clusters=32 - # ) + rewrite.check_optimization( # type: ignore[attr-defined] + model, **rewrite_specific_params if rewrite_specific_params else {} + ) if model_is_quantized and is_qat: model = rewrite.preserved_quantize(model) # type: ignore[attr-defined] checkpoints = ( @@ -501,11 +588,10 @@ def train_in_dir( loss_fn, steps_per_epoch, ) - # Placeholder for now, will be parametrized later (MLIA-1114) - # rewrite.check_optimization( # type: ignore[attr-defined] - # model, number_of_clusters=32 - # ) + rewrite.check_optimization( # type: ignore[attr-defined] + model, **rewrite_specific_params if rewrite_specific_params else {} + ) teacher.close() return output_filenames @@ -528,9 +614,13 @@ def create_model( # pylint: disable=too-many-arguments loss_fn: Callable, model_is_quantized: bool, model_to_load_from: keras.model | None = None, + rewrite_specific_params: dict | None = None, ) -> keras.Model: """Create a model, optionally from another.""" - model = rewrite(input_shape, output_shape) + if rewrite_specific_params: + model = rewrite(input_shape, output_shape, **rewrite_specific_params) + else: + model = rewrite(input_shape, output_shape) if model_is_quantized: model = rewrite.quantize(model) # type: ignore[attr-defined] model = model_compile(model, optimizer=optimizer, loss_fn=loss_fn) @@ -558,6 +648,7 @@ def model_fit( # pylint: disable=too-many-arguments loss_fn: Callable, steps_per_epoch: int, post_process: bool = False, + rewrite_specific_params: dict | None = None, ) -> keras.Model: """Train a tflite model.""" steps_so_far = 0 @@ -597,6 +688,7 @@ def model_fit( # pylint: disable=too-many-arguments loss_fn, model_is_quantized, model_to_load_from=model, + rewrite_specific_params=rewrite_specific_params, ) else: model_to_save = model diff --git a/src/mlia/nn/rewrite/library/clustering.py b/src/mlia/nn/rewrite/library/clustering.py new file mode 100644 index 0000000..9247457 --- /dev/null +++ b/src/mlia/nn/rewrite/library/clustering.py @@ -0,0 +1,79 @@ +# SPDX-FileCopyrightText: Copyright 2024, Arm Limited and/or its affiliates. +# SPDX-License-Identifier: Apache-2.0 +"""Rewrite functions used to return layers ready for clustering.""" +from typing import Any + +import tensorflow_model_optimization as tfmot +from keras.api._v2 import keras # Temporary workaround for now: MLIA-1107 + +from mlia.nn.rewrite.library.helper_functions import compute_conv2d_parameters +from mlia.nn.rewrite.library.helper_functions import get_activation_function + + +def fc_clustering_rewrite( + input_shape: Any, + output_shape: Any, + num_clusters: int = 2, + cluster_centroids_init: tfmot.clustering.keras.CentroidInitialization = tfmot.clustering.keras.CentroidInitialization( # pylint: disable=line-too-long + "CentroidInitialization.LINEAR" + ), +) -> keras.Model: + """Fully connected TensorFlow Lite model ready for clustering.""" + rewrite_params = { + "number_of_clusters": num_clusters, + "cluster_centroids_init": cluster_centroids_init, + } + model = tfmot.clustering.keras.cluster_weights( + to_cluster=keras.Sequential( + [ + keras.layers.InputLayer(input_shape=input_shape), + keras.layers.Flatten(), + keras.layers.Dense(units=output_shape), + ] + ), + **rewrite_params + ) + return model + + +def conv2d_clustering_rewrite( # pylint: disable=dangerous-default-value + input_shape: Any, + output_shape: Any, + num_clusters: int = 2, + cluster_centroids_init: tfmot.clustering.keras.CentroidInitialization = tfmot.clustering.keras.CentroidInitialization( # pylint: disable=line-too-long + "CentroidInitialization.LINEAR" + ), + activation: str = "relu", + kernel_size: list[int] = [3, 3], + layer_type: type[keras.layers.Layer] = keras.layers.Conv2D, +) -> keras.Model: + """Conv2d TensorFlow Lite model ready for clustering.""" + rewrite_params = { + "number_of_clusters": num_clusters, + "cluster_centroids_init": cluster_centroids_init, + } + conv2d_parameters = compute_conv2d_parameters( + input_shape=input_shape, + output_shape=output_shape, + kernel_size_input=kernel_size, + ) + activation_function, activation_function_extra_args = get_activation_function( + activation + ) + activation_func_found = ( # pylint: disable=duplicate-code + [activation_function(**activation_function_extra_args)] + if activation_function + else [] + ) + model = tfmot.clustering.keras.cluster_weights( + to_cluster=keras.Sequential( + [ + keras.layers.InputLayer(input_shape=input_shape), + layer_type(**conv2d_parameters), + keras.layers.BatchNormalization(), + *activation_func_found, + ] + ), + **rewrite_params + ) + return model diff --git a/src/mlia/nn/rewrite/library/fc_clustering_layer.py b/src/mlia/nn/rewrite/library/fc_clustering_layer.py deleted file mode 100644 index 7cc383e..0000000 --- a/src/mlia/nn/rewrite/library/fc_clustering_layer.py +++ /dev/null @@ -1,26 +0,0 @@ -# SPDX-FileCopyrightText: Copyright 2024, Arm Limited and/or its affiliates. -# SPDX-License-Identifier: Apache-2.0 -"""Example rewrite with one fully connected clustered layer.""" -from typing import Any - -import tensorflow_model_optimization as tfmot -from keras.api._v2 import keras # Temporary workaround for now: MLIA-1107 - - -def get_keras_model_clus(input_shape: Any, output_shape: Any) -> keras.Model: - """Generate TensorFlow Lite model for clustering rewrite.""" - rewrite_params = { - "number_of_clusters": 32, - "cluster_centroids_init": tfmot.clustering.keras.CentroidInitialization.LINEAR, - } - model = tfmot.clustering.keras.cluster_weights( - to_cluster=keras.Sequential( - [ - keras.layers.InputLayer(input_shape=input_shape), - keras.layers.Flatten(), - keras.layers.Dense(units=output_shape), - ] - ), - **rewrite_params - ) - return model diff --git a/src/mlia/nn/rewrite/library/fc_layer.py b/src/mlia/nn/rewrite/library/fc_layer.py deleted file mode 100644 index 041ce85..0000000 --- a/src/mlia/nn/rewrite/library/fc_layer.py +++ /dev/null @@ -1,18 +0,0 @@ -# SPDX-FileCopyrightText: Copyright 2023-2024, Arm Limited and/or its affiliates. -# SPDX-License-Identifier: Apache-2.0 -"""Example rewrite with one fully connected layer.""" -from typing import Any - -from keras.api._v2 import keras # Temporary workaround for now: MLIA-1107 - - -def get_keras_model(input_shape: Any, output_shape: Any) -> keras.Model: - """Generate TensorFlow Lite model for rewrite.""" - model = keras.Sequential( - ( - keras.layers.InputLayer(input_shape=input_shape), - keras.layers.Reshape([-1]), - keras.layers.Dense(output_shape), - ) - ) - return model diff --git a/src/mlia/nn/rewrite/library/fc_sparsity24_layer.py b/src/mlia/nn/rewrite/library/fc_sparsity24_layer.py deleted file mode 100644 index 531b34a..0000000 --- a/src/mlia/nn/rewrite/library/fc_sparsity24_layer.py +++ /dev/null @@ -1,23 +0,0 @@ -# SPDX-FileCopyrightText: Copyright 2024, Arm Limited and/or its affiliates. -# SPDX-License-Identifier: Apache-2.0 -"""Example rewrite with one fully connected 2:4 sparsity layer.""" -from typing import Any - -import tensorflow_model_optimization as tfmot -from keras.api._v2 import keras # Temporary workaround for now: MLIA-1107 - - -def get_keras_model(input_shape: Any, output_shape: Any) -> keras.Model: - """Generate TensorFlow Lite model for rewrite.""" - model = tfmot.sparsity.keras.prune_low_magnitude( - to_prune=keras.Sequential( - [ - keras.layers.InputLayer(input_shape=input_shape), - keras.layers.Reshape([-1]), - keras.layers.Dense(output_shape), - ] - ), - sparsity_m_by_n=(2, 4), - ) - - return model diff --git a/src/mlia/nn/rewrite/library/helper_functions.py b/src/mlia/nn/rewrite/library/helper_functions.py new file mode 100644 index 0000000..fd5993b --- /dev/null +++ b/src/mlia/nn/rewrite/library/helper_functions.py @@ -0,0 +1,61 @@ +# SPDX-FileCopyrightText: Copyright 2024, Arm Limited and/or its affiliates. +# SPDX-License-Identifier: Apache-2.0 +"""Helper functions for the rewrite library.""" +import math +from typing import Any + +import numpy as np +from keras.api._v2 import keras # Temporary workaround for now: MLIA-1107 + +ACTIVATION_FUNCTION_PRESETS = { + "relu": {"layer_func": keras.layers.ReLU, "extra_args": {}}, + "relu6": {"layer_func": keras.layers.ReLU, "extra_args": {"max_value": 6}}, + "none": {"layer_func": None, "extra_args": {}}, +} +ACTIVATION_FUNCTION_LIST = [ + act_func for act_func, _ in ACTIVATION_FUNCTION_PRESETS.items() +] + + +def get_activation_function( + activation: str = "relu", +) -> tuple[type, dict]: + """Get the activation function from a key.""" + if activation not in ACTIVATION_FUNCTION_LIST: + raise KeyError( + "Expected activation function to be " + f"in {ACTIVATION_FUNCTION_LIST}, found {activation}" + ) + activation_function = ACTIVATION_FUNCTION_PRESETS[activation]["layer_func"] + activation_function_extra_args = ACTIVATION_FUNCTION_PRESETS[activation][ + "extra_args" + ] + return activation_function, activation_function_extra_args + + +def compute_conv2d_parameters( # pylint: disable=dangerous-default-value + input_shape: np.ndarray, + output_shape: np.ndarray, + kernel_size_input: list[int] = [3, 3], +) -> dict[str, Any]: + """Compute needed kernel size and strides for a given input and output_shape.""" + input_shape = input_shape.tolist() + output_shape = output_shape.tolist() + assert len(kernel_size_input) == 2, "Kernel size should have 2 entries" + assert len(input_shape) == 3 + assert len(output_shape) == 3 + kernel_size = tuple(kernel_size_input) + num_filters = output_shape[-1] + padding = "valid" + stride_h = round(input_shape[0] / output_shape[0]) + check_output_size_h = math.floor((input_shape[0] - kernel_size[0]) / stride_h) + 1 + stride_w = round(input_shape[1] / output_shape[1]) + check_output_size_w = math.floor((input_shape[1] - kernel_size[1]) / stride_w) + 1 + if check_output_size_h != output_shape[0] or check_output_size_w != output_shape[1]: + padding = "same" + return { + "filters": num_filters, + "kernel_size": kernel_size, + "padding": padding, + "strides": (stride_h, stride_w), + } diff --git a/src/mlia/nn/rewrite/library/layers.py b/src/mlia/nn/rewrite/library/layers.py new file mode 100644 index 0000000..abf0a4c --- /dev/null +++ b/src/mlia/nn/rewrite/library/layers.py @@ -0,0 +1,53 @@ +# SPDX-FileCopyrightText: Copyright 2023-2024, Arm Limited and/or its affiliates. +# SPDX-License-Identifier: Apache-2.0 +"""Rewrite function used to return regular layers.""" +from typing import Any + +from keras.api._v2 import keras # Temporary workaround for now: MLIA-1107 + +from mlia.nn.rewrite.library.helper_functions import compute_conv2d_parameters +from mlia.nn.rewrite.library.helper_functions import get_activation_function + + +def fc_rewrite(input_shape: Any, output_shape: Any) -> keras.Model: + """Fully connected TensorFlow Lite model for rewrite.""" + model = keras.Sequential( + ( + keras.layers.InputLayer(input_shape=input_shape), + keras.layers.Reshape([-1]), + keras.layers.Dense(output_shape), + ) + ) + return model + + +def conv2d_rewrite( # pylint: disable=dangerous-default-value + input_shape: Any, + output_shape: Any, + activation: str = "relu", + kernel_size: list[int] = [3, 3], + layer_type: type[keras.layers.Layer] = keras.layers.Conv2D, +) -> keras.Model: + """Fully connected TensorFlow Lite model for rewrite.""" + conv2d_parameters = compute_conv2d_parameters( + input_shape=input_shape, + output_shape=output_shape, + kernel_size_input=kernel_size, + ) + activation_function, activation_function_extra_args = get_activation_function( + activation + ) + activation_func_found = ( # pylint: disable=duplicate-code + [activation_function(**activation_function_extra_args)] + if activation_function + else [] + ) + model = keras.Sequential( + ( + keras.layers.InputLayer(input_shape=input_shape), + layer_type(**conv2d_parameters), + keras.layers.BatchNormalization(), + *activation_func_found, + ) + ) + return model diff --git a/src/mlia/nn/rewrite/library/sparsity.py b/src/mlia/nn/rewrite/library/sparsity.py new file mode 100644 index 0000000..5102094 --- /dev/null +++ b/src/mlia/nn/rewrite/library/sparsity.py @@ -0,0 +1,146 @@ +# SPDX-FileCopyrightText: Copyright 2024, Arm Limited and/or its affiliates. +# SPDX-License-Identifier: Apache-2.0 +"""Rewrite functions used to return layers ready for sparse pruning.""" +from __future__ import annotations + +from typing import Any + +import tensorflow_model_optimization as tfmot +from keras.api._v2 import keras # Temporary workaround for now: MLIA-1107 + +from mlia.nn.rewrite.library.helper_functions import compute_conv2d_parameters +from mlia.nn.rewrite.library.helper_functions import get_activation_function + + +def fc_sparsity_unstructured_rewrite( + input_shape: Any, + output_shape: Any, + initial_sparsity: float = 0.5, + final_sparsity: float = 0.5, + begin_step: int = 0, + end_step: int = 48000, +) -> keras.Model: + """Fully connected TensorFlow Lite model ready for unstructured sparse pruning.""" + model = tfmot.sparsity.keras.prune_low_magnitude( + to_prune=keras.Sequential( + [ + keras.layers.InputLayer(input_shape=input_shape), + keras.layers.Reshape([-1]), + keras.layers.Dense(output_shape), + ] + ), + pruning_schedule=tfmot.sparsity.keras.PolynomialDecay( + initial_sparsity=initial_sparsity, + final_sparsity=final_sparsity, + begin_step=begin_step, + end_step=end_step, + ), + ) + + return model + + +def conv2d_sparsity_unstructured_rewrite( # pylint: disable=dangerous-default-value, too-many-arguments + input_shape: Any, + output_shape: Any, + initial_sparsity: float = 0.5, + final_sparsity: float = 0.5, + begin_step: int = 0, + end_step: int = 48000, + activation: str = "relu", + kernel_size: list[int] = [3, 3], + layer_type: type[keras.layers.Layer] = keras.layers.Conv2D, +) -> keras.Model: + """Conv2d TensorFlow Lite model ready for unstructured sparse pruning.""" + conv2d_parameters = compute_conv2d_parameters( + input_shape=input_shape, + output_shape=output_shape, + kernel_size_input=kernel_size, + ) + activation_function, activation_function_extra_args = get_activation_function( + activation + ) + activation_func_found = ( + [activation_function(**activation_function_extra_args)] + if activation_function + else [] + ) + model = tfmot.sparsity.keras.prune_low_magnitude( + to_prune=keras.Sequential( + [ + keras.layers.InputLayer(input_shape=input_shape), + layer_type(**conv2d_parameters), + keras.layers.BatchNormalization(), + *activation_func_found, + ] + ), + pruning_schedule=tfmot.sparsity.keras.PolynomialDecay( + initial_sparsity=initial_sparsity, + final_sparsity=final_sparsity, + begin_step=begin_step, + end_step=end_step, + ), + ) + + return model + + +def fc_sparsity_rewrite( + input_shape: Any, output_shape: Any, sparsity_m: int = 2, sparsity_n: int = 4 +) -> keras.Model: + """Fully connected TensorFlow Lite model ready for sparse pruning.""" + model = tfmot.sparsity.keras.prune_low_magnitude( + to_prune=keras.Sequential( + [ + keras.layers.InputLayer(input_shape=input_shape), + keras.layers.Reshape([-1]), + keras.layers.Dense(output_shape), + ] + ), + sparsity_m_by_n=( + sparsity_m, + sparsity_n, + ), + ) + + return model + + +def conv2d_sparsity_rewrite( # pylint: disable=dangerous-default-value + input_shape: Any, + output_shape: Any, + sparsity_m: int = 2, + sparsity_n: int = 4, + activation: str = "relu", + kernel_size: list[int] = [3, 3], + layer_type: type[keras.layers.Layer] = keras.layers.Conv2D, +) -> keras.Model: + """Conv2d TensorFlow Lite model ready for sparse pruning.""" + conv2d_parameters = compute_conv2d_parameters( + input_shape=input_shape, + output_shape=output_shape, + kernel_size_input=kernel_size, + ) + activation_function, activation_function_extra_args = get_activation_function( + activation + ) + activation_func_found = ( # pylint: disable=duplicate-code + [activation_function(**activation_function_extra_args)] + if activation_function + else [] + ) + model = tfmot.sparsity.keras.prune_low_magnitude( + to_prune=keras.Sequential( + [ + keras.layers.InputLayer(input_shape=input_shape), + layer_type(**conv2d_parameters), + keras.layers.BatchNormalization(), + *activation_func_found, + ] + ), + sparsity_m_by_n=( + sparsity_m, + sparsity_n, + ), + ) + return model diff --git a/src/mlia/nn/select.py b/src/mlia/nn/select.py index b61e713..d5470d1 100644 --- a/src/mlia/nn/select.py +++ b/src/mlia/nn/select.py @@ -17,7 +17,7 @@ from mlia.nn.common import Optimizer from mlia.nn.common import OptimizerConfiguration from mlia.nn.rewrite.core.rewrite import RewriteConfiguration from mlia.nn.rewrite.core.rewrite import RewritingOptimizer -from mlia.nn.rewrite.core.rewrite import TrainingParameters +from mlia.nn.rewrite.core.train import TrainingParameters from mlia.nn.tensorflow.config import KerasModel from mlia.nn.tensorflow.config import TFLiteModel from mlia.nn.tensorflow.optimizations.clustering import Clusterer @@ -109,7 +109,7 @@ class MultiStageOptimizer(Optimizer): def apply_optimization(self) -> None: """Apply optimization to the model.""" for config in self.optimizations: - optimizer = get_optimizer(self.model, config) + optimizer = get_optimizer(self.model, config, {}) optimizer.apply_optimization() self.model = optimizer.get_model() @@ -117,7 +117,7 @@ class MultiStageOptimizer(Optimizer): def get_optimizer( model: keras.Model | KerasModel | TFLiteModel, config: OptimizerConfiguration | OptimizationSettings | list[OptimizationSettings], - training_parameters: dict | None = None, + rewrite_parameters: dict, ) -> Optimizer: """Get optimizer for provided configuration.""" if isinstance(model, KerasModel): @@ -137,12 +137,12 @@ def get_optimizer( if isinstance(config, OptimizationSettings): return _get_optimizer( - model, cast(OptimizationSettings, config), training_parameters + model, cast(OptimizationSettings, config), rewrite_parameters ) if is_list_of(config, OptimizationSettings): return _get_optimizer( - model, cast(List[OptimizationSettings], config), training_parameters + model, cast(List[OptimizationSettings], config), rewrite_parameters ) raise ConfigurationError(f"Unknown optimization configuration {config}") @@ -151,7 +151,7 @@ def get_optimizer( def _get_optimizer( model: keras.Model | Path, optimization_settings: OptimizationSettings | list[OptimizationSettings], - training_parameters: dict | None = None, + rewrite_parameters: dict, ) -> Optimizer: if isinstance(optimization_settings, OptimizationSettings): optimization_settings = [optimization_settings] @@ -162,12 +162,12 @@ def _get_optimizer( _check_optimizer_params(opt_type, opt_target) opt_config = _get_optimizer_configuration( - opt_type, opt_target, layers_to_optimize, dataset, training_parameters + opt_type, opt_target, rewrite_parameters, layers_to_optimize, dataset ) optimizer_configs.append(opt_config) if len(optimizer_configs) == 1: - return get_optimizer(model, optimizer_configs[0]) + return get_optimizer(model, optimizer_configs[0], {}) return MultiStageOptimizer(model, optimizer_configs) @@ -189,9 +189,9 @@ def _get_rewrite_params( def _get_optimizer_configuration( optimization_type: str, optimization_target: int | float | str, + rewrite_parameters: dict, layers_to_optimize: list[str] | None = None, dataset: Path | None = None, - training_parameters: dict | None = None, ) -> OptimizerConfiguration: """Get optimizer configuration for provided parameters.""" _check_optimizer_params(optimization_type, optimization_target) @@ -212,12 +212,14 @@ def _get_optimizer_configuration( if opt_type == "rewrite": if isinstance(optimization_target, str): - rewrite_params = _get_rewrite_params(training_parameters) return RewriteConfiguration( optimization_target=str(optimization_target), layers_to_optimize=layers_to_optimize, dataset=dataset, - train_params=rewrite_params, + train_params=_get_rewrite_params(rewrite_parameters["train_params"]), + rewrite_specific_params=rewrite_parameters.get( + "rewrite_specific_params" + ), ) raise ConfigurationError( |