apply simqat on lenet

official/cv/lenet/README.md

@@ -17,6 +17,14 @@
         - [Export MindIR](#export-mindir)
         - [Infer on Ascend310](#infer-on-ascend310)
         - [result](#result)
+    - [Apply algorithm in MindSpore Golden Stick](#apply-algorithm-in-mindspore-golden-stick)
+        - [Training Process](#Training Process-1)
+            - [Running on GPU](#running-on-gpu-1)
+        - [Evaluation Process](#evaluation-process-1)
+            - [Running on GPU](#running-on-gpu-2)
+            - [Result](#resutl-3)
+        - [Inference Process](#inference-process-1)
+            - [Export MindIR](#export-mindir-1)
     - [Model Description](#model-description)
         - [Performance](#performance)
             - [Evaluation Performance](#evaluation-performance)
@@ -280,6 +288,64 @@ Inference result is saved in current path, you can find result like this in acc.
 'Accuracy': 0.9843
 ```
 
+# Apply algorithm in MindSpore Golden Stick
+
+MindSpore Golden Stick is a compression algorithm set for MindSpore. We usually apply algorithm in Golden Stick before training for smaller model size, lower power consuming or faster inference process.
+MindSpore Golden Stick provides SimQAT algorithm for Lenet5. SimQAT is a quantization-aware training algorithm that trains the quantization parameters of certain layers in the network by introducing fake-quantization nodes, so that the model can perform inference with less power consumption or higher performance during the deployment phase.
+
+## Training Process
+
+### Running on GPU
+
+```text
+cd ./golden_stick/scripts/
+# PYTHON_PATH represents path to directory of 'train.py'.
+bash run_standalone_train_gpu.sh [PYTHON_PATH] [CONFIG_FILE] [DATASET_PATH] [CKPT_TYPE](optional) [CKPT_PATH](optional)
+
+# standalone training example, apply SimQAT and train from beginning
+cd ./golden_stick/scripts/
+bash run_standalone_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/lenet_mnist_config.yaml /path/to/dataset
+
+# standalone training example, apply SimQAT and train from full precision checkpoint
+cd ./golden_stick/scripts/
+bash run_standalone_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/lenet_mnist_config.yaml /path/to/dataset FP32 /path/to/fp32_ckpt
+
+# standalone training example, apply SimQAT and train from pretrained checkpoint
+cd ./golden_stick/scripts/
+bash run_standalone_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/lenet_mnist_config.yaml /path/to/dataset PRETRAINED /path/to/pretrained_ckpt
+```
+
+## Evaluation Process
+
+### Running on GPU
+
+```text
+# evaluation
+cd ./golden_stick/scripts/
+# PYTHON_PATH represents path to directory of 'eval.py'.
+bash run_eval_gpu.sh [PYTHON_PATH] [CONFIG_FILE] [DATASET_PATH] [CHECKPOINT_PATH]
+
+# evaluation example
+cd ./golden_stick/scripts/
+bash run_eval_gpu.sh ../quantization/simqat/ ../quantization/simqat/lenet_mnist_config.yaml /path/to/dataset ./checkpoint/lenet-10.ckpt
+```
+
+### Result
+
+Evaluation result will be stored in the example path, whose folder name is "eval". Under this, you can find result like the following in log.
+
+- Apply SimQAT on Lenet5, and evaluating with MNIST dataset
+
+```bash
+================ {'Accuracy': 0.9907852564102564} ================
+```
+
+## Inference Process
+
+### Export MindIR
+
+Not support exporting MindIR now.
+
 ## [Model Description](#contents)
 
 ### [Performance](#contents)

official/cv/lenet/README_CN.md

@@ -19,6 +19,14 @@
         - [导出MindIR](#导出mindir)
         - [在Ascend310执行推理](#在ascend310执行推理)
         - [结果](#结果)
+    - [应用金箍棒模型压缩算法](#应用金箍棒模型压缩算法)
+        - [训练过程](#训练过程-1)
+            - [GPU处理器环境运行](#gpu处理器环境运行)
+        - [评估过程](#评估过程-1)
+            - [GPU处理器环境运行](#gpu处理器环境运行-1)
+            - [结果](#结果-3)
+        - [推理过程](#推理过程-1)
+            - [导出MindIR](#导出mindir-1)
     - [模型描述](#模型描述)
     - [性能](#性能)
         - [评估性能](#评估性能)
@@ -278,6 +286,68 @@ bash run_infer_310.sh [MINDIR_PATH] [DATA_PATH] [DVPP] [DEVICE_ID]
 'Accuracy':0.9843
 ```
 
+# 应用金箍棒模型压缩算法
+
+金箍棒是MindSpore的模型压缩算法集，我们可以在模型训练前应用金箍棒中的模型压缩算法，从而达到压缩模型大小、降低模型推理功耗，或者加速推理过程的目的。
+
+针对Lenet5，金箍棒提供了SimQAT算法，SimQAT是量化感知训练算法，通过引入伪量化节点来训练网络中的某些层的量化参数，从而在部署阶段，模型得以以更小的功耗或者更高的性能进行推理。
+
+## 训练过程
+
+### GPU处理器环境运行
+
+```text
+# 训练
+cd ./golden_stick/scripts/
+# PYTHON_PATH 表示需要应用的算法的'train.py'脚本所在的目录。
+bash run_standalone_train_gpu.sh [PYTHON_PATH] [CONFIG_FILE] [DATASET_PATH] [CKPT_TYPE](optional) [CKPT_PATH](optional)
+
+# 训练示例（应用SimQAT算法并从头开始量化训练）
+cd ./golden_stick/scripts/
+bash run_standalone_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/lenet_mnist_config.yaml /path/to/dataset
+
+# 训练示例（应用SimQAT算法并加载预训练的全精度checkoutpoint，并进行量化训练）
+cd ./golden_stick/scripts/
+bash run_standalone_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/lenet_mnist_config.yaml /path/to/dataset FP32 /path/to/fp32_ckpt
+
+# 训练示例（应用SimQAT算法并加载上次量化训练的checkoutpoint，继续进行量化训练）
+cd ./golden_stick/scripts/
+bash run_standalone_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/lenet_mnist_config.yaml /path/to/dataset PRETRAINED /path/to/pretrained_ckpt
+```
+
+## 评估过程
+
+### GPU处理器环境运行
+
+```text
+# 评估
+cd ./golden_stick/scripts/
+# PYTHON_PATH 表示需要应用的算法的'eval.py'脚本所在的目录。
+bash run_eval_gpu.sh [PYTHON_PATH] [CONFIG_FILE] [DATASET_PATH] [CHECKPOINT_PATH]
+```
+
+```text
+# 评估示例
+cd ./golden_stick/scripts/
+bash run_eval_gpu.sh ../quantization/simqat/ ../quantization/simqat/lenet_mnist_config.yaml /path/to/dataset ./checkpoint/lenet-10.ckpt
+```
+
+### 结果
+
+评估结果保存在示例路径中，文件夹名为“eval”。您可在此路径下的日志找到如下结果：
+
+- 使用SimQAT算法量化Lenet5，并使用MNIST数据集评估：
+
+```bash
+================ {'Accuracy': 0.9907852564102564} ================
+```
+
+## 推理过程
+
+### 导出MindIR
+
+当前暂不支持导出MindIR。
+
 ## 模型描述
 
 ## 性能

official/cv/lenet/golden_stick/quantization/simqat/algorithm.py

+# Copyright 2022 Huawei Technologies Co., Ltd
+#
+# 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.
+# ============================================================================
+"""Create SimQAT algorithm instance."""
+
+from mindspore_gs import SimulatedQuantizationAwareTraining as SimQAT
+
+
+def create_simqat():
+    algo = SimQAT()
+    algo.set_act_quant_delay(900)
+    algo.set_weight_quant_delay(900)
+    algo.set_act_symmetric(False)
+    algo.set_weight_symmetric(True)
+    algo.set_act_per_channel(False)
+    algo.set_weight_per_channel(True)
+    algo.set_enable_fusion(True)
+    algo.set_bn_fold(True)
+    algo.set_one_conv_fold(False)
+    return algo

official/cv/lenet/golden_stick/quantization/simqat/eval.py

+# Copyright 2022 Huawei Technologies Co., Ltd
+#
+# 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.
+# ============================================================================
+"""
+######################## eval lenet example ########################
+Apply quantization-aware-training algorithms on LeNet5 model and eval accuracy according to model file:
+"""
+
+import os
+from src.model_utils.config import config
+from src.dataset import create_dataset
+from src.lenet import LeNet5
+
+import mindspore.nn as nn
+from mindspore import context
+from mindspore.train.serialization import load_checkpoint, load_param_into_net
+from mindspore.train import Model
+from mindspore.nn.metrics import Accuracy
+from algorithm import create_simqat
+
+
+def eval_lenet():
+    print('eval with config: ', config)
+    if config.device_target != "GPU":
+        raise NotImplementedError("SimQAT only support running on GPU now!")
+    if config.mode_name == "GRAPH":
+        context.set_context(mode=context.GRAPH_MODE, device_target=config.device_target)
+    else:
+        context.set_context(mode=context.PYNATIVE_MODE, device_target=config.device_target)
+
+    network = LeNet5(config.num_classes)
+    # apply golden stick algorithm on LeNet5 model
+    algo = create_simqat()
+    network = algo.apply(network)
+
+    net_loss = nn.SoftmaxCrossEntropyWithLogits(sparse=True, reduction="mean")
+    model = Model(network, net_loss, metrics={"Accuracy": Accuracy()})
+
+    print("============== Starting Testing ==============")
+    param_dict = load_checkpoint(config.checkpoint_file_path)
+    load_param_into_net(network, param_dict)
+    ds_eval = create_dataset(os.path.join(config.data_path), config.batch_size)
+    if ds_eval.get_dataset_size() == 0:
+        raise ValueError("Please check dataset size > 0 and batch_size <= dataset size")
+
+    acc = model.eval(ds_eval)
+    print("============== {} ==============".format(acc))
+
+
+if __name__ == "__main__":
+    eval_lenet()

official/cv/lenet/golden_stick/quantization/simqat/lenet_mnist_config.yaml

+# Builtin Configurations(DO NOT CHANGE THESE CONFIGURATIONS unless you know exactly what you are doing)
+enable_modelarts: False
+data_url: ""
+train_url: ""
+checkpoint_url: ""
+data_path: "/cache/data"
+output_path: "/cache/train"
+device_target: GPU
+enable_profiling: False
+checkpoint_file_path: ""
+
+# ==============================================================================
+# Training options
+num_classes: 10
+lr: 0.01
+momentum: 0.9
+epoch_size: 10
+batch_size: 32
+buffer_size: 1000
+image_height: 32
+image_width: 32
+save_checkpoint_steps: 1875
+keep_checkpoint_max: 10
+device_id: 0
+file_name: "lenet-sim-qat"
+file_format: "MINDIR"
+fp32_ckpt: ""
+  
+model_name: lenet
+learning_rate: 0.002
+dataset_name: 'mnist'
+sink_size: -1
+dataset_sink_mode: True
+save_checkpoint: True
+save_checkpoint_epochs: 2
+mode_name: "GRAPH"
+
+# lenet acc calculation
+result_path: ''
+img_path: ''
+
+---
+# Config description for each option
+data_url: 'Dataset url for obs'
+train_url: 'Training output url for obs'
+data_path: 'Dataset path for local'
+output_path: 'Training output path for local'
+
+device_target: 'Target device type' 
+enable_profiling: 'Whether enable profiling while training, default: False'
+file_name: 'output file name.'
+file_format: 'file format'
+result_path: "result files path."
+img_path: "image file path."
+---
+device_target: ['GPU']
+file_format: ['AIR', 'MINDIR']

official/cv/lenet/golden_stick/quantization/simqat/train.py

+# Copyright 2022 Huawei Technologies Co., Ltd
+#
+# 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.
+# ============================================================================
+"""
+######################## train lenet example ########################
+train lenet and get network model files(.ckpt) :
+"""
+
+import os
+from src.model_utils.config import config
+from src.dataset import create_dataset
+from src.lenet import LeNet5
+
+import mindspore.nn as nn
+from mindspore import context
+from mindspore.train.callback import ModelCheckpoint, CheckpointConfig, LossMonitor, TimeMonitor
+from mindspore.train import Model
+from mindspore.nn.metrics import Accuracy
+from mindspore.common import set_seed
+from mindspore import load_checkpoint, load_param_into_net
+from algorithm import create_simqat
+
+set_seed(1)
+
+
+def train_lenet():
+    if config.device_target != "GPU":
+        raise NotImplementedError("SimQAT only support running on GPU now!")
+    if config.mode_name == "GRAPH":
+        context.set_context(mode=context.GRAPH_MODE, device_target=config.device_target)
+    else:
+        context.set_context(mode=context.PYNATIVE_MODE, device_target=config.device_target)
+    ds_train = create_dataset(os.path.join(config.data_path), config.batch_size)
+    if ds_train.get_dataset_size() == 0:
+        raise ValueError("Please check dataset size > 0 and batch_size <= dataset size")
+
+    network = LeNet5(config.num_classes)
+    if config.fp32_ckpt:
+        fp32_ckpt = load_checkpoint(config.fp32_ckpt)
+        load_param_into_net(network, fp32_ckpt)
+    # apply golden stick algorithm on LeNet5 model
+    algo = create_simqat()
+    network = algo.apply(network)
+
+    net_loss = nn.SoftmaxCrossEntropyWithLogits(sparse=True, reduction="mean")
+    net_opt = nn.Momentum(network.trainable_params(), config.lr, config.momentum)
+    time_cb = TimeMonitor(data_size=ds_train.get_dataset_size())
+    config_ck = CheckpointConfig(save_checkpoint_steps=config.save_checkpoint_steps,
+                                 keep_checkpoint_max=config.keep_checkpoint_max)
+    ckpoint_cb = ModelCheckpoint(prefix="checkpoint_lenet", directory="./ckpt", config=config_ck)
+
+    context.set_context(enable_graph_kernel=True)
+    model = Model(network, net_loss, net_opt, metrics={"Accuracy": Accuracy()})
+
+    print("============== Starting Training ==============")
+    model.train(config.epoch_size, ds_train, callbacks=[time_cb, ckpoint_cb, LossMonitor()])
+
+
+if __name__ == "__main__":
+    train_lenet()

official/cv/lenet/golden_stick/scripts/run_eval_gpu.sh

+#!/bin/bash
+# Copyright 2022 Huawei Technologies Co., Ltd
+#
+# 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.
+# ============================================================================
+
+# an simple tutorial as follows, more parameters can be setting
+CURPATH="$(dirname "$0")"
+
+if [ $# != 4 ]
+then
+    echo "Usage: bash run_eval_gpu.sh [PYTHON_PATH] [CONFIG_FILE] [DATASET_PATH] [CHECKPOINT_PATH]"
+    echo "PYTHON_PATH represents path to directory of 'eval.py'."
+    exit 1
+fi
+
+get_real_path(){
+  if [ "${1:0:1}" == "/" ]; then
+    echo "$1"
+  else
+    echo "$(realpath -m $PWD/$1)"
+  fi
+}
+
+PYTHON_PATH=$(get_real_path $1)
+CONFIG_FILE=$(get_real_path $2)
+DATASET_PATH=$(get_real_path $3)
+CKPT_FILE=$(get_real_path $4)
+
+
+if [ ! -d $PYTHON_PATH ]
+then
+    echo "error: PYTHON_PATH=$PYTHON_PATH is not a directory"
+exit 1
+fi
+
+if [ ! -f $CONFIG_FILE ]
+then
+    echo "error: CONFIG_FILE=$CONFIG_FILE is not a file"
+exit 1
+fi
+
+if [ ! -d $DATASET_PATH ]
+then
+    echo "error: DATASET_PATH=$DATASET_PATH is not a directory"
+exit 1
+fi
+
+if [ ! -f $CKPT_FILE ]
+then
+    echo "error: CKPT_FILE=$CKPT_FILE is not a file"
+exit 1
+fi
+
+ulimit -u unlimited
+export DEVICE_NUM=1
+export DEVICE_ID=0
+export RANK_SIZE=$DEVICE_NUM
+export RANK_ID=0
+
+if [ -d "eval" ];
+then
+    rm -rf ./eval
+fi
+mkdir ./eval
+cp ${PYTHON_PATH}/*.py ./eval
+cp -r ${CURPATH}/../../src ./eval
+cd ./eval || exit
+env > env.log
+echo "start evaluation for device $DEVICE_ID"
+python eval.py --config_path=$CONFIG_FILE --data_path=$DATASET_PATH --device_target="GPU" --checkpoint_file_path=$CKPT_FILE > log.txt 2>&1 &
+cd ..

official/cv/lenet/golden_stick/scripts/run_standalone_train_gpu.sh

+#!/bin/bash
+# Copyright 2022 Huawei Technologies Co., Ltd
+#
+# 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.
+# ============================================================================
+
+CURPATH="$(dirname "$0")"
+# shellcheck source=/dev/null
+
+if [ $# != 3 ] && [ $# != 5 ]; then
+    echo "Usage: bash run_standalone_train_gpu.sh [PYTHON_PATH] [CONFIG_FILE] [DATASET_PATH] [CKPT_TYPE](optional) [CKPT_PATH](optional)"
+    echo "Examples:"
+    echo "  Train from the beginning:"
+    echo "    bash run_standalone_train_gpu.sh /path/to/train.py lenet_config.yaml /path/to/dataset"
+    echo "  Train from full precision checkpoint:"
+    echo "    bash run_standalone_train_gpu.sh /path/to/train.py lenet_config.yaml /path/to/dataset FP32 /path/to/fp32_ckpt"
+    echo "  Train from pretrained checkpoint:"
+    echo "    bash run_standalone_train_gpu.sh /path/to/train.py lenet_config.yaml /path/to/dataset PRETRAINED /path/to/pretrained_ckpt"
+    exit 1
+fi
+
+get_real_path(){
+  if [ "${1:0:1}" == "/" ]; then
+    echo "$1"
+  else
+    echo "$(realpath -m $PWD/$1)"
+  fi
+}
+
+PYTHON_PATH=$(get_real_path $1)
+CONFIG_FILE=$(get_real_path $2)
+DATASET_PATH=$(get_real_path $3)
+
+if [ ! -d $PYTHON_PATH ]
+then
+    echo "error: PYTHON_PATH=$PYTHON_PATH is not a directory"
+    exit 1
+fi
+
+if [ ! -f $CONFIG_FILE ]
+then
+    echo "error: CONFIG_FILE=$CONFIG_FILE is not a file"
+    exit 1
+fi
+
+if [ ! -d $DATASET_PATH ]
+then
+    echo "error: DATASET_PATH=$DATASET_PATH is not a directory"
+    exit 1
+fi
+
+if [ $# == 5 ]; then
+  CKPT_TYPE=$4
+  CKPT_FILE=$(get_real_path $5)
+
+  if [ "x$CKPT_TYPE" != "xFP32" ] && [ "x$CKPT_TYPE" != "xPRETRAINED" ]; then
+      echo "error: CKPT_TYPE=$CKPT_TYPE is not valid, should be FP32 or PRETRAINED"
+      exit 1
+  fi
+  if [ ! -f $CKPT_FILE ]; then
+      echo "error: CKPT_FILE=$CKPT_FILE is not a file"
+      exit 1
+  fi
+fi
+
+ulimit -u unlimited
+export DEVICE_NUM=1
+export RANK_SIZE=1
+
+if [ -d "train" ];
+then
+    rm -rf ./train
+fi
+mkdir ./train
+cp ${PYTHON_PATH}/*.py ./train
+cp -r ${CURPATH}/../../src ./train
+cd ./train || exit
+
+if [ "x$CKPT_TYPE" == "xFP32" ]; then
+  python train.py --config_path=$CONFIG_FILE --device_target="GPU" --data_path=$DATASET_PATH \
+          --fp32_ckpt=$CKPT_FILE &> log &
+elif [ "x$CKPT_TYPE" == "xPRETRAINED" ]; then
+  python train.py --config_path=$CONFIG_FILE --device_target="GPU" --data_path=$DATASET_PATH \
+          --pre_trained=$CKPT_FILE &> log &
+else
+  python train.py --config_path=$CONFIG_FILE --device_target="GPU" --data_path=$DATASET_PATH &> log &
+fi

official/cv/resnet/README.md

@@ -30,15 +30,13 @@
         - [Infer on Ascend310](#infer-on-ascend310)
         - [result](#result-2)
 - [Apply algorithm in MindSpore Golden Stick](#apply-algorithm-in-mindspore-golden-stick)
-    - [Apply SimQAT algorithm](#apply-simqat-algorithm)
-        - [Training Process](#Training Process-1)
-            - [Running on GPU](#running-on-gpu-2)
-        - [Resume Process](#resume-process-1)
-        - [Evaluation Process](#evaluation-process-1)
-            - [Running on GPU](#running-on-gpu-3)
-            - [Result](#resutl-3)
-        - [Inference Process](#inference-process-1)
-            - [Export MindIR](#export-mindir-1)
+    - [Training Process](#Training Process-1)
+        - [Running on GPU](#running-on-gpu-2)
+    - [Evaluation Process](#evaluation-process-1)
+        - [Running on GPU](#running-on-gpu-3)
+        - [Result](#resutl-3)
+    - [Inference Process](#inference-process-1)
+        - [Export MindIR](#export-mindir-1)
 - [Model Description](#model-description)
     - [Performance](#performance)
         - [Evaluation Performance](#evaluation-performance)
@@ -811,41 +809,52 @@ Total data: 50000, top1 accuracy: 0.76844, top5 accuracy: 0.93522.
 
 # Apply algorithm in MindSpore Golden Stick
 
-## Apply SimQAT algorithm
+MindSpore Golden Stick is a compression algorithm set for MindSpore. We usually apply algorithm in Golden Stick before training for smaller model size, lower power consuming or faster inference process.
+MindSpore Golden Stick provides SimQAT algorithm for ResNet50. SimQAT is a quantization-aware training algorithm that trains the quantization parameters of certain layers in the network by introducing fake-quantization nodes, so that the model can perform inference with less power consumption or higher performance during the deployment phase.
 
-SimQAT is a quantization-aware training algorithm that trains the quantization parameters of certain layers in the network by introducing pseudo-quantization nodes, so that the model can perform inference with less power consumption or higher performance during the deployment phase.
+## Training Process
 
-### Training Process
-
-#### Running on GPU
+### Running on GPU
 
 ```text
 # distributed training
 cd ./golden_stick/scripts/
 # PYTHON_PATH represents path to directory of 'train.py'.
-bash run_distribute_train_gpu.sh [PYTHON_PATH] [CONFIG_FILE] [DATASET_PATH] [FP32_CKPT_PATH]
+bash run_distribute_train_gpu.sh [PYTHON_PATH] [CONFIG_FILE] [DATASET_PATH] [CKPT_TYPE](optional) [CKPT_PATH](optional)
 
-# distributed training example
+# distributed training example, apply SimQAT and train from beginning
+cd ./golden_stick/scripts/
+bash run_distribute_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/resnet50_cifar10_config.yaml ./cifar10/train/
+
+# distributed training example, apply SimQAT and train from full precision checkpoint
+cd ./golden_stick/scripts/
+bash run_distribute_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/resnet50_cifar10_config.yaml /path/to/dataset FP32 /path/to/fp32_ckpt
+
+# distributed training example, apply SimQAT and train from pretrained checkpoint
 cd ./golden_stick/scripts/
-bash run_distribute_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/resnet50_cifar10_config.yaml ./cifar10/train/ ./checkpoint/resnet-90.ckpt
+bash run_distribute_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/resnet50_cifar10_config.yaml /path/to/dataset PRETRAINED /path/to/pretrained_ckpt
 
 # standalone training
 cd ./golden_stick/scripts/
 # PYTHON_PATH represents path to directory of 'train.py'.
-bash run_standalone_train_gpu.sh [PYTHON_PATH] [CONFIG_FILE] [DATASET_PATH] [FP32_CKPT_PATH]
+bash run_standalone_train_gpu.sh [PYTHON_PATH] [CONFIG_FILE] [DATASET_PATH] [CKPT_TYPE](optional) [CKPT_PATH](optional)
 
-# standalone training example
+# standalone training example, apply SimQAT and train from beginning
 cd ./golden_stick/scripts/
-bash run_standalone_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/resnet50_cifar10_config.yaml ./cifar10/train/ ./checkpoint/resnet-90.ckpt
-```
+bash run_standalone_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/resnet50_cifar10_config.yaml ./cifar10/train/
 
-### Resume Process
+# standalone training example, apply SimQAT and train from full precision checkpoint
+cd ./golden_stick/scripts/
+bash run_standalone_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/resnet50_cifar10_config.yaml /path/to/dataset FP32 /path/to/fp32_ckpt
 
-SimQAT algorithm is not support to resume training process now.
+# standalone training example, apply SimQAT and train from pretrained checkpoint
+cd ./golden_stick/scripts/
+bash run_standalone_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/resnet50_cifar10_config.yaml /path/to/dataset PRETRAINED /path/to/pretrained_ckpt
+```
 
-### Evaluation Process
+## Evaluation Process
 
-#### Running on GPU
+### Running on GPU
 
 ```text
 # evaluation
@@ -858,19 +867,19 @@ cd ./golden_stick/scripts/
 bash run_eval_gpu.sh ../quantization/simqat/ ../quantization/simqat/resnet50_cifar10_config.yaml ./cifar10/train/ ./checkpoint/resnet-90.ckpt
 ```
 
-#### Result
+### Result
 
 Evaluation result will be stored in the example path, whose folder name is "eval". Under this, you can find result like the following in log.
 
-- Apply SumQAT on ResNet50, and evaluating with CIFAR-10 dataset
+- Apply SimQAT on ResNet50, and evaluating with CIFAR-10 dataset
 
 ```bash
 result:{'top_1_accuracy': 0.9354967948717948, 'top_5_accuracy': 0.9981971153846154} ckpt=~/resnet50_cifar10/train_parallel0/resnet-180_195.ckpt
 ```
 
-### Inference Process
+## Inference Process
 
-#### Export MindIR
+### Export MindIR
 
 Not support exporting MindIR now.
 

official/cv/resnet/README_CN.md

@@ -32,15 +32,14 @@
         - [在Ascend310执行推理](#在ascend310执行推理)
         - [结果](#结果-2)
 - [应用金箍棒模型压缩算法](#应用金箍棒模型压缩算法)
-    - [应用SimQAT算法](#应用simqat算法)
-        - [训练过程](#训练过程-1)
-            - [GPU处理器环境运行](#gpu处理器环境运行-2)
-        - [续训过程](#续训过程-1)
-        - [评估过程](#评估过程-1)
-            - [GPU处理器环境运行](#gpu处理器环境运行-3)
-            - [结果](#结果-3)
-        - [推理过程](#推理过程-1)
-            - [导出MindIR](#导出mindir-1)
+    - [训练过程](#训练过程-1)
+        - [GPU处理器环境运行](#gpu处理器环境运行-2)
+    - [续训过程](#续训过程-1)
+    - [评估过程](#评估过程-1)
+        - [GPU处理器环境运行](#gpu处理器环境运行-3)
+        - [结果](#结果-3)
+    - [推理过程](#推理过程-1)
+        - [导出MindIR](#导出mindir-1)
 - [模型描述](#模型描述)
     - [性能](#性能)
         - [评估性能](#评估性能)
@@ -770,46 +769,58 @@ Total data: 50000, top1 accuracy: 0.76844, top5 accuracy: 0.93522.
 
 # 应用金箍棒模型压缩算法
 
-## 应用SimQAT算法
+金箍棒是MindSpore的模型压缩算法集，我们可以在模型训练前应用金箍棒中的模型压缩算法，从而达到压缩模型大小、降低模型推理功耗，或者加速推理过程的目的。
 
-SimQAT是一种量化感知训练算法，通过引入伪量化节点来训练网络中的某些层的量化参数，从而在部署阶段，模型得以以更小的功耗或者更高的性能进行推理。
+针对ResNet50，金箍棒提供了SimQAT算法，SimQAT是一种量化感知训练算法，通过引入伪量化节点来训练网络中的某些层的量化参数，从而在部署阶段，模型得以以更小的功耗或者更高的性能进行推理。
 
-### 训练过程
+## 训练过程
 
-#### GPU处理器环境运行
+### GPU处理器环境运行
 
 ```text
 # 分布式训练
 cd ./golden_stick/scripts/
-# PYTHON_PATH 表示'train.py'脚本所在的目录。
-bash run_distribute_train_gpu.sh [PYTHON_PATH] [CONFIG_FILE] [DATASET_PATH] [FP32_CKPT_PATH]
+# PYTHON_PATH 表示需要应用的算法的'train.py'脚本所在的目录。
+bash run_distribute_train_gpu.sh [PYTHON_PATH] [CONFIG_FILE] [DATASET_PATH] [CKPT_TYPE](optional) [CKPT_PATH](optional)
 
-# 分布式训练示例
+# 分布式训练示例（应用SimQAT算法并从头开始量化训练）
+cd ./golden_stick/scripts/
+bash run_distribute_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/resnet50_cifar10_config.yaml ./cifar10/train/
+
+# 分布式训练示例（应用SimQAT算法并加载预训练的全精度checkoutpoint，进行量化训练）
+cd ./golden_stick/scripts/
+bash run_distribute_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/resnet50_cifar10_config.yaml /path/to/dataset FP32 /path/to/fp32_ckpt
+
+# 分布式训练示例（应用SimQAT算法并加载之前训练的checkoutpoint，继续进行量化训练）
 cd ./golden_stick/scripts/
-bash run_distribute_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/resnet50_cifar10_config.yaml ./cifar10/train/ ./checkpoint/resnet-90.ckpt
+bash run_distribute_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/resnet50_cifar10_config.yaml /path/to/dataset PRETRAINED /path/to/pretrained_ckpt
 
 # 单机训练
 cd ./golden_stick/scripts/
-# PYTHON_PATH 表示'train.py'脚本所在的目录。
-bash run_standalone_train_gpu.sh [PYTHON_PATH] [CONFIG_FILE] [DATASET_PATH] [FP32_CKPT_PATH]
+# PYTHON_PATH 表示需要应用的算法的'train.py'脚本所在的目录。
+bash run_standalone_train_gpu.sh [PYTHON_PATH] [CONFIG_FILE] [DATASET_PATH] [CKPT_TYPE](optional) [CKPT_PATH](optional)
 
-# 单机训练示例
+# 单机训练示例（应用SimQAT算法并从头开始量化训练）
 cd ./golden_stick/scripts/
-bash run_standalone_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/resnet50_cifar10_config.yaml ./cifar10/train/ ./checkpoint/resnet-90.ckpt
-```
+bash run_standalone_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/resnet50_cifar10_config.yaml ./cifar10/train/
 
-### 续训过程
+# 单机训练示例（应用SimQAT算法并加载预训练的全精度checkoutpoint，并进行量化训练）
+cd ./golden_stick/scripts/
+bash run_standalone_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/resnet50_cifar10_config.yaml /path/to/dataset FP32 /path/to/fp32_ckpt
 
-SimQAT算法当前暂不支持续训。
+# 单机训练示例（应用SimQAT算法并加载上次量化训练的checkoutpoint，继续进行量化训练）
+cd ./golden_stick/scripts/
+bash run_standalone_train_gpu.sh ../quantization/simqat/ ../quantization/simqat/resnet50_cifar10_config.yaml /path/to/dataset PRETRAINED /path/to/pretrained_ckpt
+```
 
-### 评估过程
+## 评估过程
 
-#### GPU处理器环境运行
+### GPU处理器环境运行
 
 ```text
 # 评估
 cd ./golden_stick/scripts/
-# PYTHON_PATH 表示'eval.py'脚本所在的目录。
+# PYTHON_PATH 表示需要应用的算法的'eval.py'脚本所在的目录。
 bash run_eval_gpu.sh [PYTHON_PATH] [CONFIG_FILE] [DATASET_PATH] [CHECKPOINT_PATH]
 ```
 
@@ -819,7 +830,7 @@ cd ./golden_stick/scripts/
 bash run_eval_gpu.sh ../quantization/simqat/ ../quantization/simqat/resnet50_cifar10_config.yaml ./cifar10/train/ ./checkpoint/resnet-90.ckpt
 ```
 
-#### 结果
+### 结果
 
 评估结果保存在示例路径中，文件夹名为“eval”。您可在此路径下的日志找到如下结果：
 
@@ -829,9 +840,9 @@ bash run_eval_gpu.sh ../quantization/simqat/ ../quantization/simqat/resnet50_cif
 result:{'top_1_accuracy': 0.9354967948717948, 'top_5_accuracy': 0.9981971153846154} ckpt=~/resnet50_cifar10/train_parallel0/resnet-180_195.ckpt
 ```
 
-### 推理过程
+## 推理过程
 
-#### 导出MindIR
+### 导出MindIR
 
 当前暂不支持导出MindIR。
 

official/cv/resnet/golden_stick/quantization/simqat/resnet50_cifar10_config.yaml

@@ -9,7 +9,6 @@ run_distribute: False
 enable_profiling: False
 data_path: "/cache/data"
 output_path: "/cache/train"
-load_path: "/cache/checkpoint_path/"
 device_target: "Ascend"
 checkpoint_path: "./checkpoint/"
 checkpoint_file_path: ""
@@ -17,7 +16,6 @@ checkpoint_file_path: ""
 # ==============================================================================
 # Training options
 optimizer: "Momentum"
-infer_label: ""
 class_num: 10
 batch_size: 32
 loss_scale: 1024
@@ -40,15 +38,8 @@ dataset: "cifar10"
 device_num: 1
 pre_trained: ""
 fp32_ckpt: ""
-run_eval: False
-eval_dataset_path: ""
-parameter_server: False
 filter_weight: False
 save_best_ckpt: True
-eval_start_epoch: 40
-eval_interval: 1
-enable_cache: False
-cache_session_id: ""
 mode_name: "GRAPH"
 boost_mode: "O0"
 conv_init: "XavierUniform"
@@ -76,20 +67,13 @@ save_graphs_path: "./graphs"
 has_trained_epoch: 0
 has_trained_step: 0
 
-# postprocess resnet inference
-result_path: ''
-label_path: ''
-
 ---
 # Help description for each configuration
-enable_modelarts: "Whether training on modelarts, default: False"
 data_url: "Dataset url for obs"
 checkpoint_url: "The location of checkpoint for obs"
 data_path: "Dataset path for local"
 output_path: "Training output path for local"
-load_path: "The location of checkpoint for obs"
 device_target: "Target device type, available: [Ascend, GPU, CPU]"
-enable_profiling: "Whether enable profiling while training, default: False"
 num_classes: "Class for dataset"
 batch_size: "Batch size for training and evaluation"
 epoch_size: "Total training epochs."