NPU (HUAWEI Ascend)¶

Usage¶

Please refer to the building documentation of MMCV to install MMCV and mmengine on NPU devices.

Here we use 8 NPUs on your computer to train the model with the following command:

bash ./tools/dist_train.sh configs/resnet/resnet50_8xb32_in1k.py 8

Also, you can use only one NPU to train the model with the following command:

python ./tools/train.py configs/resnet/resnet50_8xb32_in1k.py

Model	Top-1 (%)	Top-5 (%)	Config	Download
ResNet-50	76.40	93.21	config	model \| log
ResNetXt-32x4d-50	77.48	93.75	config	model \| log
HRNet-W18	77.06	93.57	config	model \| log
ResNetV1D-152	79.41	94.48	config	model \| log
SE-ResNet-50	77.65	93.74	config	model \|log
ShuffleNetV2 1.0x	69.52	88.79	config	model \| log
MobileNetV2	71.74	90.28	config	model \| log
MobileNetV3-Small	67.09	87.17	config	model \| log
*CSPResNeXt50	77.25	93.46	config	model \| log
*EfficientNet-B4	75.73	92.91	config	model \|log
**DenseNet121	72.53	90.85	config	model \| log

Notes:

If not specially marked, the results are almost same between results on the NPU and results on the GPU with FP32.
(*) The training results of these models are lower than the results on the readme in the corresponding model, mainly because the results on the readme are directly the weight of the timm of the eval, and the results on this side are retrained according to the config with mmcls. The results of the config training on the GPU are consistent with the results of the NPU.
(**) The accuracy of this model is slightly lower because config is a 4-card config, we use 8 cards to run, and users can adjust hyperparameters to get the best accuracy results.

All above models are provided by Huawei Ascend group.