-
Notifications
You must be signed in to change notification settings - Fork 258
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Add keras saved_model examples (#1201)
- Loading branch information
1 parent
e81b265
commit fdd40e1
Showing
45 changed files
with
2,288 additions
and
3 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
33 changes: 33 additions & 0 deletions
33
...w/image_recognition/keras_models/inception_resnet_v2/quantization/ptq/README.md
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,33 @@ | ||
| Step-by-Step | ||
| ============ | ||
|
|
||
| This document is used to enable Tensorflow Keras models using Intel® Neural Compressor. | ||
|
|
||
|
|
||
| ## Prerequisite | ||
|
|
||
| ### 1. Installation | ||
| ```shell | ||
| # Install Intel® Neural Compressor | ||
| pip install neural-compressor | ||
| ``` | ||
| ### 2. Install Intel Tensorflow | ||
| ```shell | ||
| pip install intel-tensorflow | ||
| ``` | ||
| > Note: Supported Tensorflow [Version](../../../../../../../README.md). | ||
| ### 3. Prepare Pretrained model | ||
|
|
||
| The pretrained model is provided by [Keras Applications](https://keras.io/api/applications/). prepare the model, Run as follow: | ||
| ``` | ||
| python prepare_model.py --output_model=/path/to/model | ||
| ``` | ||
| `--output_model ` the model should be saved as SavedModel format or H5 format. | ||
|
|
||
| ## Run Command | ||
| ```shell | ||
| bash run_tuning.sh --config=inception_resnet_v2.yaml --input_model=./path/to/model --output_model=./result --eval_data=/path/to/evaluation/dataset --calib_data=/path/to/calibration/dataset | ||
| bash run_benchmark.sh --config=inception_resnet_v2.yaml --input_model=./path/to/model --mode=performance --eval_data=/path/to/evaluation/dataset | ||
| ``` | ||
|
|
42 changes: 42 additions & 0 deletions
42
...ge_recognition/keras_models/inception_resnet_v2/quantization/ptq/inception_resnet_v2.yaml
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,42 @@ | ||
| # | ||
| # Copyright (c) 2021 Intel Corporation | ||
| # | ||
| # Licensed under the Apache License, Version 2.0 (the "License"); | ||
| # you may not use this file except in compliance with the License. | ||
| # You may obtain a copy of the License at | ||
| # | ||
| # http://www.apache.org/licenses/LICENSE-2.0 | ||
| # | ||
| # Unless required by applicable law or agreed to in writing, software | ||
| # distributed under the License is distributed on an "AS IS" BASIS, | ||
| # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| # See the License for the specific language governing permissions and | ||
| # limitations under the License. | ||
|
|
||
| version: 1.0 | ||
|
|
||
| model: # mandatory. used to specify model specific information. | ||
| name: inception_resnet_v2 | ||
| framework: tensorflow # mandatory. supported values are tensorflow, pytorch, pytorch_ipex, onnxrt_integer, onnxrt_qlinear or mxnet; allow new framework backend extension. | ||
|
|
||
| quantization: # optional. tuning constraints on model-wise for advance user to reduce tuning space. | ||
| calibration: | ||
| sampling_size: 50, 100 # optional. default value is 100. used to set how many samples should be used in calibration. | ||
| model_wise: # optional. tuning constraints on model-wise for advance user to reduce tuning space. | ||
| activation: | ||
| algorithm: minmax | ||
|
|
||
| evaluation: # optional. required if user doesn't provide eval_func in neural_compressor.Quantization. | ||
| accuracy: | ||
| performance: # optional. used to benchmark performance of passing model. | ||
| iteration: 100 | ||
| configs: | ||
| cores_per_instance: 4 | ||
| num_of_instance: 7 | ||
|
|
||
| tuning: | ||
| accuracy_criterion: | ||
| relative: 0.01 # optional. default value is relative, other value is absolute. this example allows relative accuracy loss: 1%. | ||
| exit_policy: | ||
| timeout: 0 # optional. tuning timeout (seconds). default value is 0 which means early stop. combine with max_trials field to decide when to exit. | ||
| random_seed: 9527 # optional. random seed for deterministic tuning. |
128 changes: 128 additions & 0 deletions
128
...es/tensorflow/image_recognition/keras_models/inception_resnet_v2/quantization/ptq/main.py
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,128 @@ | ||
| # | ||
| # -*- coding: utf-8 -*- | ||
| # | ||
| # Copyright (c) 2018 Intel Corporation | ||
| # | ||
| # Licensed under the Apache License, Version 2.0 (the "License"); | ||
| # you may not use this file except in compliance with the License. | ||
| # You may obtain a copy of the License at | ||
| # | ||
| # http://www.apache.org/licenses/LICENSE-2.0 | ||
| # | ||
| # Unless required by applicable law or agreed to in writing, software | ||
| # distributed under the License is distributed on an "AS IS" BASIS, | ||
| # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| # See the License for the specific language governing permissions and | ||
| # limitations under the License. | ||
| # | ||
| import time | ||
| import shutil | ||
| import numpy as np | ||
| from argparse import ArgumentParser | ||
| from neural_compressor import data | ||
| import tensorflow as tf | ||
| tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR) | ||
|
|
||
| flags = tf.compat.v1.flags | ||
| FLAGS = flags.FLAGS | ||
|
|
||
| ## Required parameters | ||
| flags.DEFINE_string( | ||
| 'input_model', None, 'Run inference with specified pb graph.') | ||
|
|
||
| flags.DEFINE_string( | ||
| 'output_model', None, 'The output model of the quantized model.') | ||
|
|
||
| flags.DEFINE_string( | ||
| 'mode', 'performance', 'define benchmark mode for accuracy or performance') | ||
|
|
||
| flags.DEFINE_bool( | ||
| 'tune', False, 'whether to tune the model') | ||
|
|
||
| flags.DEFINE_bool( | ||
| 'benchmark', False, 'whether to benchmark the model') | ||
|
|
||
| flags.DEFINE_string( | ||
| 'config', 'bert.yaml', 'yaml configuration of the model') | ||
|
|
||
| flags.DEFINE_string( | ||
| 'calib_data', None, 'location of calibration dataset') | ||
|
|
||
| flags.DEFINE_string( | ||
| 'eval_data', None, 'location of evaluate dataset') | ||
|
|
||
| from neural_compressor.experimental.metric.metric import TensorflowTopK | ||
| from neural_compressor.experimental.data.transforms.transform import ComposeTransform | ||
| from neural_compressor.experimental.data.datasets.dataset import TensorflowImageRecord | ||
| from neural_compressor.experimental.data.transforms.imagenet_transform import LabelShift | ||
| from neural_compressor.experimental.data.dataloaders.default_dataloader import DefaultDataLoader | ||
| from neural_compressor.data.transforms.imagenet_transform import BilinearImagenetTransform | ||
|
|
||
| eval_dataset = TensorflowImageRecord(root=FLAGS.eval_data, transform=ComposeTransform(transform_list= \ | ||
| [BilinearImagenetTransform(height=299, width=299)])) | ||
| if FLAGS.benchmark and FLAGS.mode == 'performance': | ||
| eval_dataloader = DefaultDataLoader(dataset=eval_dataset, batch_size=1) | ||
| else: | ||
| eval_dataloader = DefaultDataLoader(dataset=eval_dataset, batch_size=32) | ||
| if FLAGS.calib_data: | ||
| calib_dataset = TensorflowImageRecord(root=FLAGS.calib_data, transform=ComposeTransform(transform_list= \ | ||
| [BilinearImagenetTransform(height=299, width=299)])) | ||
| calib_dataloader = DefaultDataLoader(dataset=calib_dataset, batch_size=10) | ||
|
|
||
| def evaluate(model, measurer=None): | ||
| """ | ||
| Custom Evaluate function to inference the model for specified metric on validation dataset. | ||
| Args: | ||
| model ([tf.saved_model.load]): The model will be the class of tf.saved_model.load(quantized_model_path). | ||
| measurer (object, optional): for precise benchmark measurement. | ||
| Returns: | ||
| [float]: evaluation result, the larger is better. | ||
| """ | ||
| infer = model.signatures["serving_default"] | ||
| output_dict_keys = infer.structured_outputs.keys() | ||
| output_name = list(output_dict_keys )[0] | ||
| postprocess = LabelShift(label_shift=1) | ||
| metric = TensorflowTopK(k=1) | ||
|
|
||
| def eval_func(dataloader, metric): | ||
| results = [] | ||
| for idx, (inputs, labels) in enumerate(dataloader): | ||
| inputs = np.array(inputs) | ||
| input_tensor = tf.constant(inputs) | ||
| if measurer: | ||
| measurer.start() | ||
| predictions = infer(input_tensor)[output_name] | ||
| if measurer: | ||
| measurer.end() | ||
| predictions = predictions.numpy() | ||
| predictions, labels = postprocess((predictions, labels)) | ||
| metric.update(predictions, labels) | ||
| return results | ||
|
|
||
| results = eval_func(eval_dataloader, metric) | ||
| acc = metric.result() | ||
| return acc | ||
|
|
||
| def main(_): | ||
| if FLAGS.tune: | ||
| from neural_compressor.experimental import Quantization, common | ||
| quantizer = Quantization(FLAGS.config) | ||
| quantizer.model = common.Model(FLAGS.input_model) | ||
| quantizer.eval_func = evaluate | ||
| quantizer.calib_dataloader = calib_dataloader | ||
| q_model = quantizer.fit() | ||
| q_model.save(FLAGS.output_model) | ||
|
|
||
|
|
||
| if FLAGS.benchmark: | ||
| from neural_compressor.experimental import Benchmark, common | ||
| evaluator = Benchmark(FLAGS.config) | ||
| evaluator.model = common.Model(FLAGS.input_model) | ||
| evaluator.b_func = evaluate | ||
| evaluator.b_dataloader = eval_dataloader | ||
| evaluator(FLAGS.mode) | ||
|
|
||
| if __name__ == "__main__": | ||
| tf.compat.v1.app.run() |
17 changes: 17 additions & 0 deletions
17
...flow/image_recognition/keras_models/inception_resnet_v2/quantization/ptq/prepare_model.py
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,17 @@ | ||
| import argparse | ||
| import tensorflow as tf | ||
| def get_inception_resnet_v2_model(saved_path): | ||
| model = tf.keras.applications.InceptionResNetV2(weights='imagenet') | ||
| model.save(saved_path) | ||
|
|
||
| if __name__ == "__main__": | ||
| parser = argparse.ArgumentParser( | ||
| description='Export pretained keras model', | ||
| formatter_class=argparse.ArgumentDefaultsHelpFormatter) | ||
| parser.add_argument( | ||
| '--output_model', | ||
| type=str, | ||
| help='path to exported model file') | ||
|
|
||
| args = parser.parse_args() | ||
| get_inception_resnet_v2_model(args.output_model) |
44 changes: 44 additions & 0 deletions
44
...flow/image_recognition/keras_models/inception_resnet_v2/quantization/ptq/run_benchmark.sh
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,44 @@ | ||
| #!/bin/bash | ||
| set -x | ||
|
|
||
| function main { | ||
|
|
||
| init_params "$@" | ||
| run_benchmark | ||
|
|
||
| } | ||
|
|
||
| # init params | ||
| function init_params { | ||
| for var in "$@" | ||
| do | ||
| case $var in | ||
| --config=*) | ||
| config=$(echo $var |cut -f2 -d=) | ||
| ;; | ||
| --input_model=*) | ||
| input_model=$(echo $var |cut -f2 -d=) | ||
| ;; | ||
| --mode=*) | ||
| mode=$(echo $var |cut -f2 -d=) | ||
| ;; | ||
| --eval_data=*) | ||
| eval_data=$(echo $var |cut -f2 -d=) | ||
| ;; | ||
| esac | ||
| done | ||
|
|
||
| } | ||
|
|
||
| # run_tuning | ||
| function run_benchmark { | ||
|
|
||
| python main.py \ | ||
| --input_model ${input_model} \ | ||
| --config ${config} \ | ||
| --benchmark \ | ||
| --mode ${mode} \ | ||
| --eval_data ${eval_data} | ||
| } | ||
|
|
||
| main "$@" |
Oops, something went wrong.