[南京理工大学][高校贡献][Pytorch][Convmixer]--初次提交 · Pull Request !389 · Ascend/ModelZoo-PyTorch

# Convmixer: Patches Are All You Need?
https://github.com/locuslab/convmixer

## Requirements

- Install PyTorch ([pytorch.org](http://pytorch.org)), torch >= 1.4.0, torchvision >= 0.5.0
- `pip install -r requirements.txt`
- Download the ImageNet dataset from http://www.image-net.org/

## Training the convmixer

To train a model, run `train_npu.py` with the desired model architecture and the path to the ImageNet dataset:

training on 8 npu
```
bash ./test/train_full_8p.sh  --data_path=xxx

```

get model performance
```
1. test 1p performance  
bash ./test/train_performance_1p.sh  --data_path=xxx

2. test 8p performance  
bash ./test/train_performance_8p.sh  --data_path=xxx   
```

## Validation
```
bash ./test/train_eval_1p.sh  --data_path=xxx --checkpoint=XXX
```

## Traing log
```
test/output/${device_id}/train_${device_id}.log              # training detail log

test/output/${device_id}/${Network}_bs${BatchSize}_${RANK_SIZE}'p'_'acc'.log            # training accuracy result

```

## Convmixer  training result
| name | Acc@1   |  FPS  | Npu nums | Epochs  | AMP_Type | s/ per step  |
| :--: | :-----: | :---: | :------: | :-----: | :------: | :---------:  |
| NPU  |   -     | 50.24 |      1   |   150   |    O2    |    1.284     |
| NPU  | 80.298% | 406.18|      8   |   150   |    O2    |    1.257     |

Ascend/ModelZoo-PyTorch .gitee-modal { width: 500px !important; }

搜索帮助

Ascend/ModelZoo-PyTorch