【Not Merge】support resnet50 train cifar10

README.md

@@ -1,10 +1,10 @@
-# Train CIFAR10 with OneFlow
+# Train CIFAR10 with OneFlow Cambricon in MLU270
 
-I'm playing with [OneFlow](https://github.com/Oneflow-Inc/oneflow) on the CIFAR10 dataset.
+I'm playing with [oneflow-cambricon](https://github.com/Oneflow-Inc/oneflow-cambricon) on the CIFAR10 dataset.
 
 ## Prerequisites
-- Python 3.6+
-- OneFlow 0.5.0rc+
+- Python 3.7+
+- OneFlow cambricon
 
 ## Training
 ```
@@ -18,55 +18,7 @@ python main.py --resume --lr=0.01
 ## Accuracy
 | Model             | Acc.        |
 | ----------------- | ----------- |
-| [VGG16](https://arxiv.org/abs/1409.1556)              | 93.92%|
-| [ResNet18](https://arxiv.org/abs/1512.03385)          | 95.62%|
-| [ResNet50](https://arxiv.org/abs/1512.03385)          | 95.40% |
-| [ResNet101](https://arxiv.org/abs/1512.03385)         |       |
-| [RegNetX_200MF](https://arxiv.org/abs/2003.13678)     | 95.10%|
-| [RegNetY_400MF](https://arxiv.org/abs/2003.13678)     |       |
-| [MobileNetV2](https://arxiv.org/abs/1801.04381)       | 92.56%|
-| [ResNeXt29(32x4d)](https://arxiv.org/abs/1611.05431)  |       |
-| [ResNeXt29(2x64d)](https://arxiv.org/abs/1611.05431)  |       |
-| [SimpleDLA](https://arxiv.org/abs/1707.064)           |       |
-| [DenseNet121](https://arxiv.org/abs/1608.06993)       |       |
-| [PreActResNet18](https://arxiv.org/abs/1603.05027)    |       |
-| [DPN92](https://arxiv.org/abs/1707.01629)             |       |
-| [DLA](https://arxiv.org/pdf/1707.06484.pdf)           |       |
-
-## Quantization Aware Training
-
-If you are interested in OneFlow FX feature, please do the following to compile OneFlow Experience FX.
-
-```
-git clone https://github.com/Oneflow-Inc/oneflow
-cd oneflow
-git checkout add_fx_intermediate_representation
-mkdir build
-cd build
-cmake -DCUDNN_ROOT_DIR=/usr/local/cudnn -DCMAKE_BUILD_TYPE=Release -DTHIRD_PARTY_MIRROR=aliyun -DUSE_CLANG_FORMAT=ON -DTREAT_WARNINGS_AS_ERRORS=OFF ..
-make -j32
-```
-
-```
-# Start training with: 
-python main_qat.py
-
-# You can manually resume the training with: 
-python main_qat.py --resume --lr=0.01
-```
-
-Note:
-
-The `momentum` parameter in the `MovingAverageMinMaxObserver` class defaults to 0.95, which will not be changed in the following experiments. 
-## Accuracy
-| Model             | quantization_bit | quantization_scheme | quantization_formula | per_layer_quantization | Acc |
-| ----------------- | ----------- | ----------- | ----------- | ----------- | ----------- |
-| ResNet18          |  8     |  symmetric      | google       |   True     |  95.19%      | 
-| ResNet18          |  8     |  symmetric      | google       |   False    |  95.24%      | 
-| ResNet18          |  8     |  affine         | google       |   True     |  95.32%      | 
-| ResNet18          |  8     |  affine         | google       |   False    |  95.30%      | 
-| ResNet18          |  8     |  symmetric      | cambricon    |   True     |  95.19%      |
-
+| [ResNet50](https://arxiv.org/abs/1512.03385)          |  |
 ## Reference
 - https://github.com/kuangliu/pytorch-cifar
 

main.py

@@ -3,10 +3,8 @@
 import oneflow.nn as nn
 import oneflow.optim as optim
 import oneflow.nn.functional as F
-import oneflow.backends.cudnn as cudnn
 
-import torch
-import torchvision
+import flowvision
 
 
 import os
@@ -22,7 +20,7 @@
                     help='resume from checkpoint')
 args = parser.parse_args()
 
-device = 'cuda' if flow.cuda.is_available() else 'cpu'
+device = 'mlu'
 best_acc = 0  # best test accuracy
 start_epoch = 0  # start from epoch 0 or last checkpoint epoch
 
@@ -54,7 +52,7 @@
 #     testset, batch_size=100, shuffle=False, num_workers=2)
 
 # PyTorch DataReader
-import torchvision.transforms as transforms
+import flowvision.transforms as transforms
 transform_train = transforms.Compose([
     transforms.RandomCrop(32, padding=4),
     transforms.RandomHorizontalFlip(),
@@ -67,14 +65,14 @@
     transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),
 ])
 
-trainset = torchvision.datasets.CIFAR10(
+trainset = flowvision.datasets.CIFAR10(
     root='./data', train=True, download=True, transform=transform_train)
-trainloader = torch.utils.data.DataLoader(
+trainloader = flow.utils.data.DataLoader(
     trainset, batch_size=128, shuffle=True, num_workers=2, drop_last=True)
 
-testset = torchvision.datasets.CIFAR10(
+testset = flowvision.datasets.CIFAR10(
     root='./data', train=False, download=True, transform=transform_test)
-testloader = torch.utils.data.DataLoader(
+testloader = flow.utils.data.DataLoader(
     testset, batch_size=100, shuffle=False, num_workers=2, drop_last=True)
 
 
@@ -84,8 +82,8 @@
 # Model
 print('==> Building model..')
 # net = VGG('VGG16')
-net = ResNet18()
-# net = ResNet50()
+# net = ResNet18()
+net = ResNet50()
 # net = PreActResNet18()
 # net = GoogLeNet()
 # net = DenseNet121()
@@ -145,8 +143,8 @@ def train(epoch):
         # _, predicted = outputs.max(1)
         predicted = flow.argmax(outputs, 1).to(flow.int64)
         total += targets.size(0)
-
-        correct += predicted.eq(targets).to(flow.int32).sum().item()
+        tmp = predicted.eq(targets).to(flow.int)
+        correct += tmp.sum().item()
 
         progress_bar(batch_idx, len(trainloader), 'Loss: %.3f | Acc: %.3f%% (%d/%d)'
                      % (train_loss/(batch_idx+1), 100.*correct/total, correct, total))
@@ -175,7 +173,7 @@ def test(epoch):
             total += targets.size(0)
 
 
-            correct += predicted.eq(targets).to(flow.int32).sum().item()
+            correct += predicted.eq(targets).to(flow.int).sum().item()
 
             progress_bar(batch_idx, len(testloader), 'Loss: %.3f | Acc: %.3f%% (%d/%d)'
                          % (test_loss/(batch_idx+1), 100.*correct/total, correct, total))