Source code for aimet_onnx.layer_output_utils

# -*- mode: python -*-
# =============================================================================
#  @@-COPYRIGHT-START-@@
#
#  Copyright (c) 2023, Qualcomm Innovation Center, Inc. All rights reserved.
#
#  Redistribution and use in source and binary forms, with or without
#  modification, are permitted provided that the following conditions are met:
#
#  1. Redistributions of source code must retain the above copyright notice,
#     this list of conditions and the following disclaimer.
#
#  2. Redistributions in binary form must reproduce the above copyright notice,
#     this list of conditions and the following disclaimer in the documentation
#     and/or other materials provided with the distribution.
#
#  3. Neither the name of the copyright holder nor the names of its contributors
#     may be used to endorse or promote products derived from this software
#     without specific prior written permission.
#
#  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
#  AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
#  IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
#  ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
#  LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
#  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
#  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
#  INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
#  CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
#  ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
#  POSSIBILITY OF SUCH DAMAGE.
#
#  SPDX-License-Identifier: BSD-3-Clause
#
#  @@-COPYRIGHT-END-@@
# =============================================================================

""" This module contains utilities to capture and save intermediate layer-outputs of a model """

import copy
from typing import List, Dict, Tuple, Union
import re
import numpy as np
import onnxruntime as ort
import onnx
from packaging import version

# pylint: disable=wrong-import-order
from aimet_common.utils import AimetLogger
from aimet_common.layer_output_utils import SaveInputOutput, save_layer_output_names

from aimet_onnx.quantsim import QuantizationSimModel
from aimet_onnx.utils import create_input_dict, add_hook_to_get_activation

# pylint: disable=no-name-in-module, ungrouped-imports
if version.parse(onnx.__version__) >= version.parse("1.14.0"):
    from onnx import ModelProto
else:
    from onnx.onnx_pb import ModelProto

logger = AimetLogger.get_area_logger(AimetLogger.LogAreas.LayerOutputs)


[docs]class LayerOutputUtil: """ Implementation to capture and save outputs of intermediate layers of a model (fp32/quantsim) """ def __init__(self, model: ModelProto, dir_path: str, device: int = 0): """ Constructor - It initializes the utility classes that captures and saves layer-outputs :param model: ONNX model :param dir_path: Directory wherein layer-outputs will be saved :param device: CUDA device-id to be used """ self.model = model # Fetch appropriate execution providers depending on availability providers = ['CPUExecutionProvider'] if 'CUDAExecutionProvider' in ort.get_available_providers(): providers = [('CUDAExecutionProvider', {'device_id': device}), 'CPUExecutionProvider'] # Utility to capture layer-outputs self.layer_output = LayerOutput(model=model, providers=providers, dir_path=dir_path) # Utility to save model inputs and their corresponding layer-outputs self.save_input_output = SaveInputOutput(dir_path, 'NCHW')
[docs] def generate_layer_outputs(self, input_batch: Union[np.ndarray, List[np.ndarray], Tuple[np.ndarray]]): """ This method captures output of every layer of a model & saves the inputs and corresponding layer-outputs to disk. :param input_batch: Batch of inputs for which we want to obtain layer-outputs. :return: None """ logger.info("Generating layer-outputs for %d input instances", len(input_batch)) input_dict = create_input_dict(self.model, input_batch) layer_output_dict = self.layer_output.get_outputs(input_dict) self.save_input_output.save(input_batch, layer_output_dict) logger.info('Layer-outputs generated for %d input instances', len(input_batch))
class LayerOutput: """ This class creates a layer-output name to layer-output dictionary. """ def __init__(self, model: ModelProto, providers: List, dir_path: str): """ Constructor - It initializes few lists that are required for capturing and naming layer-outputs. :param model: ONNX model :param providers: execution providers to execute onnxruntime :param dir_path: directory to store topological order of layer-output names """ self.model = copy.deepcopy(model) self.activation_names = LayerOutput.get_activation_names(self.model) quantized_activation_names = [name for name in self.activation_names if name.endswith('_updated')] if quantized_activation_names: self.activation_names = quantized_activation_names LayerOutput.register_activations(self.model, self.activation_names) self.session = QuantizationSimModel.build_session(self.model, providers) # Replace special characters with underscore. This gives valid file names to store activation tensors. self.sanitized_activation_names = [re.sub(r'\W+', "_", name.replace('_updated', '')) for name in self.activation_names] # Save activation names which are in topological order of model graph. This order can be used while comparing layer-outputs. save_layer_output_names(self.sanitized_activation_names, dir_path) def get_outputs(self, input_dict: Dict) -> Dict[str, np.ndarray]: """ This function creates layer-output name to layer-output dictionary. :param input_dict: input name to input tensor map :return: layer-output name to layer-output dictionary """ activation_values = self.session.run(self.activation_names, input_dict) return dict(zip(self.sanitized_activation_names, activation_values)) @staticmethod def get_activation_names(model: ModelProto) -> List[str]: """ This function fetches the activation names (layer-output names) of the given onnx model. :param model: ONNX model :return: list of activation names """ activation_names = [] constant_activations = [] # No need to capture them for node in model.graph.input: activation_names.append(node.name) for node in model.graph.node: if node.op_type == 'Constant': # Ignore and keep track to further ignore its quantized version constant_activations.append(node.output[0]+'_updated') else: for output in node.output: if output not in constant_activations: activation_names.append(output) return activation_names @staticmethod def register_activations(model: ModelProto, activation_names: List): """ This function adds the intermediate activations into the model's ValueInfoProto so that they can be fetched via running the session. :param model: ONNX model :param activation_names: list of activation names to be registered :return: """ for act_name in activation_names: _ = add_hook_to_get_activation(model, act_name)