ImportError: cannot import name 'ConvBNReLUFusion' from 'torch.quantization.fx.fusion_patterns'

[huangweiwh]

When I try to ptq mobilenetv2, I get an error: "ImportError: cannot import name 'ConvBNReLUFusion' from 'torch.quantization.fx.fusion_patterns'"

[corin-lyz]

Have you solved this problem? I also encountered this problem

[wuchangli]

torch==1.10.0 has this, but the newest torch removed it. How do adapt it to the newest torch?

[ardeal]

I am able to solve this problem.

1. if you install torch 1.10.0, you are not able to install other related packages.
2. if you run MQBench on the latest torch, you are not able to import those mudules as which is removed by Pytorch.

I have already update this module and run it on the latest torch. The quantization accuracy is also good.

[bieyl]

@ardeal ImportError: cannot import name 'ConvBNReLUFusion' from 'torch.quantization.fx.fusion_patterns'
How do you sovle it? My torch version is 2.4.0, but i need to use the corresponding CUDA. So do you heve the good sollution?

[bieyl]

I am able to solve this problem.

1. if you install torch 1.10.0, you are not able to install other related packages.
2. if you run MQBench on the latest torch, you are not able to import those mudules as which is removed by Pytorch.

I have already update this module and run it on the latest torch. The quantization accuracy is also good.

Hello @ardeal ImportError: cannot import name 'ConvBNReLUFusion' from 'torch.quantization.fx.fusion_patterns'
How do you sovle it? My torch version is 2.4.0, but i need to use the corresponding CUDA. So do you heve the good sollution?