MNN MNN转换qat模型出错

我在使用mnnconvert转换qat模型时，出现如下错误：

Start to Optimize the MNN Net...
**Tensor shape**: 3, 1, 640, 640,
Error for compute convolution shape, inputCount:3, outputCount:16, KH:3, KW:3, group:1
inputChannel: 1, batch:3, width:640, height:640. Input data channel may be mismatch with filter channel count
**Tensor shape**: 3, 1, 640, 640,
Error for compute convolution shape, inputCount:3, outputCount:16, KH:3, KW:3, group:1
inputChannel: 1, batch:3, width:640, height:640. Input data channel may be mismatch with filter channel count
inputTensors : [ images, ]
outputTensors: [ outputs, ]
Converted Success!

虽然转换成功了，但是在运行时，再次出现如下错误：

The device supports: i8sdot:1, fp16:1, i8mm: 0, sve2: 0, sme2: 0
**Tensor shape**: 3, 1, 640, 640,
Error for compute convolution shape, inputCount:3, outputCount:16, KH:3, KW:3, group:1
inputChannel: 1, batch:3, width:640, height:640. Input data channel may be mismatch with filter channel count
Compute Shape Error for /backbone/stem/rbr_reparam/Conv_output_0
Can't run session because not resized