Torch2trt error - number of kernel weights does not match tensor dimensions

kmarx · March 4, 2023, 1:26am

When using torch2trt to convert a Torch model to TensorRT, I get the following errors:

/usr/local/lib/python3.8/dist-packages/torch2trt/dataset.py:61: SyntaxWarning: assertion is always true, perhaps remove parentheses?
  assert(len(self) > 0, 'Cannot create default flattener without input data.')
/home/prakashlab/.local/lib/python3.8/site-packages/torchvision/io/image.py:13: UserWarning: Failed to load image Python extension:
  warn(f"Failed to load image Python extension: {e}")

I then see the following errors:

[03/01/2023-12:20:00] [TRT] [E] 3: (Unnamed Layer* 147) [Convolution]:kernel weights has count 45 but 225 was expected
[03/01/2023-12:20:00] [TRT] [E] 4: (Unnamed Layer* 147) [Convolution]: count of 45 weights in kernel, but kernel dimensions (3,3) with 5 input channels, 5 output channels and 1 groups were specified. Expected Weights count is 5 * 3*3 * 5 / 1 = 225
[03/01/2023-12:20:00] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 147) [Convolution]: number of kernel weights does not match tensor dimensions)

There are a few layers that give the following error code:

[helpers.cpp::getHW::32] Error Code 2: Internal Error (Assertion d.nbDims >= 2 failed. )

I also get similar errors for the other layers, not just Unnamed Layer* 147. The model runs fine using torch; I am not sure why the conversion does not work.

Here’s information about my environment:

torch                  1.13.0a0+936e9305.nv22.11
torch2trt              0.4.0
torchvision            0.14.0
tensorrt               8.4.1.5
Build cuda_11.4.r11.4/compiler.31294910_0

I am trying to use TensorRT to deploy M2U-Net (GitHub - laibe/M2U-Net) to my Jetson Xavier NX. The model is defined in m2unet.py in this repository. I am trying to convert this model to TensorRT. I am using the following Python code to convert:

import torch
import torch2trt
from m2unet import M2UNet, Encoder

def main():
    modelpath = "torch_model.pth"
    savepath = "tensorrt_model.trt"
    NCH = 1
    BATCHSZ = 1
    im_sz = 1024
    sz = (BATCHSZ, NCH, im_sz, im_sz) # size of input - batch, channels, x, y

    # define your PyTorch model and load the weights
    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
    encoder = Encoder(NUM_CHAN=NCH)
    model = M2UNet(encoder, NUM_CHAN=NCH).float().to(device).cuda().eval().half()
    model.load_state_dict(torch.load(modelpath))

    # create an example input
    input_data = torch.ones(sz).cuda().half()
    # convert the PyTorch model to a TensorRT engine
    engine = torch2trt.torch2trt(model, [input_data], fp16_mode=True)
    # save the engine to a file
    torch.save(engine.state_dict(), savepath)

if __name__ == "__main__":
    main()

AastaLLL · March 6, 2023, 4:40am

Hi,

Could you share the complete log so we can get more information about this issue?

Thanks.

kmarx · March 6, 2023, 5:42pm

Hi,

It also looks like I got a few details in my original post wrong - I am using a slightly modified model but my point about it running fine with torch and getting the error stands.

Here’s the code for the model:


import torch
import torch.nn as nn
import math
def conv_bn(inp, oup, stride):
    return nn.Sequential(
        nn.Conv2d(inp, oup, 3, stride, 1, bias=False),
        nn.BatchNorm2d(oup),
        nn.ReLU(inplace=True)
    )


def conv_1x1_bn(inp, oup):
    return nn.Sequential(
        nn.Conv2d(inp, oup, 1, 1, 0, bias=False),
        nn.BatchNorm2d(oup),
        nn.ReLU(inplace=True)
    )

class InvertedResidual(nn.Module):
    def __init__(self, inp, oup, stride, expand_ratio):
        super(InvertedResidual, self).__init__()
        self.stride = stride
        self.add = nn.quantized.FloatFunctional()

        
        assert stride in [1, 2]
        
        # self.quant = torch.quantization.QuantStub()
        # self.dequant = torch.quantization.DeQuantStub()


        hidden_dim = round(inp * expand_ratio)
        self.use_res_connect = self.stride == 1 and inp == oup

        if expand_ratio == 1:
            # depthwise separable convolution block
            self.conv = nn.Sequential(
                # dw
                nn.Conv2d(hidden_dim, hidden_dim, 3, stride, 1, groups=hidden_dim, bias=False),
                nn.BatchNorm2d(hidden_dim),
                nn.ReLU(inplace=True),
                # pw-linear
                nn.Conv2d(hidden_dim, oup, 1, 1, 0, bias=False),
                nn.BatchNorm2d(oup),
            )
        else:
            # Bottleneck with expansion layer
            self.conv = nn.Sequential(
                # pw
                nn.Conv2d(inp, hidden_dim, 1, 1, 0, bias=False),
                nn.BatchNorm2d(hidden_dim),
                nn.ReLU(inplace=True),
                # dw
                nn.Conv2d(hidden_dim, hidden_dim, 3, stride, 1, groups=hidden_dim, bias=False),
                nn.BatchNorm2d(hidden_dim),
                nn.ReLU(inplace=True),
                # pw-linear
                nn.Conv2d(hidden_dim, oup, 1, 1, 0, bias=False),
                nn.BatchNorm2d(oup),
            )

    def forward(self, x):
        if self.use_res_connect:
            # return x + self.conv(x)
            return self.add.add(x, self.conv(x))
        else:
            return self.conv(x)
        
class Encoder(nn.Module):
    """
    14 layers of MobileNetv2 as encoder part
    """
    def __init__(self, NUM_CHAN=1):
        super(Encoder, self).__init__()
        block = InvertedResidual
        interverted_residual_setting = [
            # t, c, n, s
            [1, 16, 1, 1],
            [6, 24, 2, 2],
            [6, 32, 3, 2],
            [6, 64, 4, 2],
            [6, 96, 3, 1],
        ]
        # Encoder Part
        input_channel = 32 # number of input channels to first inverted (residual) block
        self.layers = [conv_bn(NUM_CHAN, 32, 2)]
        # building inverted residual blocks
        for t, c, n, s in interverted_residual_setting:
            output_channel = c
            for i in range(n):
                if i == 0:
                    self.layers.append(block(input_channel, output_channel, s, expand_ratio=t))
                else:
                    self.layers.append(block(input_channel, output_channel, 1, expand_ratio=t))
                input_channel = output_channel
        # make it nn.Sequential
        self.layers = nn.Sequential(*self.layers)
                
class DecoderBlock(nn.Module):
    """
    Decoder block: upsample and concatenate with features maps from the encoder part
    """
    def __init__(self,up_in_c,x_in_c,upsamplemode='bilinear',expand_ratio=0.15):
        super().__init__()
        self.cat = nn.quantized.FloatFunctional()

        self.upsample = nn.Upsample(scale_factor=2,mode=upsamplemode,align_corners=False) # H, W -> 2H, 2W
        self.ir1 = InvertedResidual(up_in_c+x_in_c,(x_in_c + up_in_c) // 2,stride=1,expand_ratio=expand_ratio)

    def forward(self,up_in,x_in):
        up_out = self.upsample(up_in)
        # cat_x = torch.cat([up_out, x_in] , dim=1)
        cat_x = self.cat.cat([up_out, x_in] , dim=1)
        x = self.ir1(cat_x)
        return x
    
class LastDecoderBlock(nn.Module):
    def __init__(self,x_in_c,upsamplemode='bilinear',expand_ratio=0.15, output_channels=1, activation='linear'):
        super().__init__()
        self.upsample = nn.Upsample(scale_factor=2,mode=upsamplemode,align_corners=False) # H, W -> 2H, 2W
        self.ir1 = InvertedResidual(x_in_c,16,stride=1,expand_ratio=expand_ratio)
        layers =  [
            nn.Conv2d(16, output_channels, 1, 1, 0, bias=True),
        ]
        self.cat = nn.quantized.FloatFunctional()

        if activation == 'sigmoid':
            layers.append(nn.Sigmoid())
        elif activation == 'softmax':
            layers.append(nn.Softmax(dim=1))
        elif activation == 'linear' or activation is None:
            pass
        else:
            raise NotImplementedError('Activation {} not implemented'.format(activation))
        self.conv = nn.Sequential(
           *layers
        )

    def forward(self,up_in,x_in):
        up_out = self.upsample(up_in)
        # cat_x = torch.cat([up_out, x_in] , dim=1)
        cat_x = self.cat.cat([up_out, x_in] , dim=1)
        x = self.ir1(cat_x)
        x = self.conv(x)
        return x
    
class M2UNet(nn.Module):
        def __init__(self,encoder,NUM_CHAN=1,upsamplemode='bilinear',output_channels=1, activation="linear", expand_ratio=0.15):
            super(M2UNet,self).__init__()
            encoder = list(encoder.children())[0]
            # Encoder
            self.conv1 = encoder[0:2]
            self.conv2 = encoder[2:4]
            self.conv3 = encoder[4:7]
            self.conv4 = encoder[7:14]
            # Decoder
            self.decode4 = DecoderBlock(96,32,upsamplemode,expand_ratio)
            self.decode3 = DecoderBlock(64,24,upsamplemode,expand_ratio)
            self.decode2 = DecoderBlock(44,16,upsamplemode,expand_ratio)
            self.decode1 = LastDecoderBlock(30 + NUM_CHAN,upsamplemode,expand_ratio, output_channels=output_channels, activation=activation)
            # initilaize weights 
            self._init_params()

        def _init_params(self):
            for m in self.modules():
                if isinstance(m, nn.Conv2d):
                    n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
                    m.weight.data.normal_(0, math.sqrt(2. / n))
                    if m.bias is not None:
                        m.bias.data.zero_()
                elif isinstance(m, nn.BatchNorm2d):
                    m.weight.data.fill_(1)
                    m.bias.data.zero_()
                elif isinstance(m, nn.Linear):
                    n = m.weight.size(1)
                    m.weight.data.normal_(0, 0.01)
                    m.bias.data.zero_()
            
        
        
        def forward(self,x):
            conv1 = self.conv1(x)
            conv2 = self.conv2(conv1)
            conv3 = self.conv3(conv2)
            conv4 = self.conv4(conv3)
            decode4 = self.decode4(conv4,conv3)
            decode3 = self.decode3(decode4,conv2)
            decode2 = self.decode2(decode3,conv1)
            decode1 = self.decode1(decode2,x)
            return decode1

Here’s the error in full:

/usr/local/lib/python3.8/dist-packages/torch2trt/dataset.py:61: SyntaxWarning: assertion is always true, perhaps remove parentheses?
  assert(len(self) > 0, 'Cannot create default flattener without input data.')
/home/prakashlab/.local/lib/python3.8/site-packages/torchvision/io/image.py:13: UserWarning: Failed to load image Python extension: 
  warn(f"Failed to load image Python extension: {e}")
successful load
[03/06/2023-09:32:42] [TRT] [E] 3: (Unnamed Layer* 3) [Convolution]:kernel weights has count 288 but 9216 was expected
[03/06/2023-09:32:42] [TRT] [E] 4: (Unnamed Layer* 3) [Convolution]: count of 288 weights in kernel, but kernel dimensions (3,3) with 32 input channels, 32 output channels and 1 groups were specified. Expected Weights count is 32 * 3*3 * 32 / 1 = 9216
[03/06/2023-09:32:42] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 3) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:42] [TRT] [E] 3: (Unnamed Layer* 3) [Convolution]:kernel weights has count 288 but 9216 was expected
[03/06/2023-09:32:42] [TRT] [E] 4: (Unnamed Layer* 3) [Convolution]: count of 288 weights in kernel, but kernel dimensions (3,3) with 32 input channels, 32 output channels and 1 groups were specified. Expected Weights count is 32 * 3*3 * 32 / 1 = 9216
[03/06/2023-09:32:42] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 3) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:42] [TRT] [E] 3: (Unnamed Layer* 3) [Convolution]:kernel weights has count 288 but 9216 was expected
[03/06/2023-09:32:42] [TRT] [E] 4: (Unnamed Layer* 3) [Convolution]: count of 288 weights in kernel, but kernel dimensions (3,3) with 32 input channels, 32 output channels and 1 groups were specified. Expected Weights count is 32 * 3*3 * 32 / 1 = 9216
[03/06/2023-09:32:42] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 3) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:42] [TRT] [E] 2: [helpers.cpp::getHW::32] Error Code 2: Internal Error (Assertion d.nbDims >= 2 failed. )
[03/06/2023-09:32:42] [TRT] [E] 3: (Unnamed Layer* 11) [Convolution]:kernel weights has count 864 but 82944 was expected
[03/06/2023-09:32:42] [TRT] [E] 4: (Unnamed Layer* 11) [Convolution]: count of 864 weights in kernel, but kernel dimensions (3,3) with 96 input channels, 96 output channels and 1 groups were specified. Expected Weights count is 96 * 3*3 * 96 / 1 = 82944
[03/06/2023-09:32:43] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 11) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:43] [TRT] [E] 3: (Unnamed Layer* 11) [Convolution]:kernel weights has count 864 but 82944 was expected
[03/06/2023-09:32:43] [TRT] [E] 4: (Unnamed Layer* 11) [Convolution]: count of 864 weights in kernel, but kernel dimensions (3,3) with 96 input channels, 96 output channels and 1 groups were specified. Expected Weights count is 96 * 3*3 * 96 / 1 = 82944
[03/06/2023-09:32:43] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 11) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:43] [TRT] [E] 3: (Unnamed Layer* 11) [Convolution]:kernel weights has count 864 but 82944 was expected
[03/06/2023-09:32:43] [TRT] [E] 4: (Unnamed Layer* 11) [Convolution]: count of 864 weights in kernel, but kernel dimensions (3,3) with 96 input channels, 96 output channels and 1 groups were specified. Expected Weights count is 96 * 3*3 * 96 / 1 = 82944
[03/06/2023-09:32:43] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 11) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:43] [TRT] [E] 2: [helpers.cpp::getHW::32] Error Code 2: Internal Error (Assertion d.nbDims >= 2 failed. )
[03/06/2023-09:32:43] [TRT] [E] 3: (Unnamed Layer* 19) [Convolution]:kernel weights has count 1296 but 186624 was expected
[03/06/2023-09:32:43] [TRT] [E] 4: (Unnamed Layer* 19) [Convolution]: count of 1296 weights in kernel, but kernel dimensions (3,3) with 144 input channels, 144 output channels and 1 groups were specified. Expected Weights count is 144 * 3*3 * 144 / 1 = 186624
[03/06/2023-09:32:43] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 19) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:43] [TRT] [E] 3: (Unnamed Layer* 19) [Convolution]:kernel weights has count 1296 but 186624 was expected
[03/06/2023-09:32:43] [TRT] [E] 4: (Unnamed Layer* 19) [Convolution]: count of 1296 weights in kernel, but kernel dimensions (3,3) with 144 input channels, 144 output channels and 1 groups were specified. Expected Weights count is 144 * 3*3 * 144 / 1 = 186624
[03/06/2023-09:32:43] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 19) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:43] [TRT] [E] 3: (Unnamed Layer* 19) [Convolution]:kernel weights has count 1296 but 186624 was expected
[03/06/2023-09:32:43] [TRT] [E] 4: (Unnamed Layer* 19) [Convolution]: count of 1296 weights in kernel, but kernel dimensions (3,3) with 144 input channels, 144 output channels and 1 groups were specified. Expected Weights count is 144 * 3*3 * 144 / 1 = 186624
[03/06/2023-09:32:43] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 19) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:43] [TRT] [E] 2: [helpers.cpp::getHW::32] Error Code 2: Internal Error (Assertion d.nbDims >= 2 failed. )
[03/06/2023-09:32:43] [TRT] [E] 3: (Unnamed Layer* 28) [Convolution]:kernel weights has count 1296 but 186624 was expected
[03/06/2023-09:32:43] [TRT] [E] 4: (Unnamed Layer* 28) [Convolution]: count of 1296 weights in kernel, but kernel dimensions (3,3) with 144 input channels, 144 output channels and 1 groups were specified. Expected Weights count is 144 * 3*3 * 144 / 1 = 186624
[03/06/2023-09:32:43] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 28) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:43] [TRT] [E] 3: (Unnamed Layer* 28) [Convolution]:kernel weights has count 1296 but 186624 was expected
[03/06/2023-09:32:43] [TRT] [E] 4: (Unnamed Layer* 28) [Convolution]: count of 1296 weights in kernel, but kernel dimensions (3,3) with 144 input channels, 144 output channels and 1 groups were specified. Expected Weights count is 144 * 3*3 * 144 / 1 = 186624
[03/06/2023-09:32:43] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 28) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:43] [TRT] [E] 3: (Unnamed Layer* 28) [Convolution]:kernel weights has count 1296 but 186624 was expected
[03/06/2023-09:32:43] [TRT] [E] 4: (Unnamed Layer* 28) [Convolution]: count of 1296 weights in kernel, but kernel dimensions (3,3) with 144 input channels, 144 output channels and 1 groups were specified. Expected Weights count is 144 * 3*3 * 144 / 1 = 186624
[03/06/2023-09:32:44] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 28) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:44] [TRT] [E] 2: [helpers.cpp::getHW::32] Error Code 2: Internal Error (Assertion d.nbDims >= 2 failed. )
[03/06/2023-09:32:44] [TRT] [E] 3: (Unnamed Layer* 36) [Convolution]:kernel weights has count 1728 but 331776 was expected
[03/06/2023-09:32:44] [TRT] [E] 4: (Unnamed Layer* 36) [Convolution]: count of 1728 weights in kernel, but kernel dimensions (3,3) with 192 input channels, 192 output channels and 1 groups were specified. Expected Weights count is 192 * 3*3 * 192 / 1 = 331776
[03/06/2023-09:32:44] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 36) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:44] [TRT] [E] 3: (Unnamed Layer* 36) [Convolution]:kernel weights has count 1728 but 331776 was expected
[03/06/2023-09:32:44] [TRT] [E] 4: (Unnamed Layer* 36) [Convolution]: count of 1728 weights in kernel, but kernel dimensions (3,3) with 192 input channels, 192 output channels and 1 groups were specified. Expected Weights count is 192 * 3*3 * 192 / 1 = 331776
[03/06/2023-09:32:44] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 36) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:44] [TRT] [E] 3: (Unnamed Layer* 36) [Convolution]:kernel weights has count 1728 but 331776 was expected
[03/06/2023-09:32:44] [TRT] [E] 4: (Unnamed Layer* 36) [Convolution]: count of 1728 weights in kernel, but kernel dimensions (3,3) with 192 input channels, 192 output channels and 1 groups were specified. Expected Weights count is 192 * 3*3 * 192 / 1 = 331776
[03/06/2023-09:32:44] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 36) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:44] [TRT] [E] 2: [helpers.cpp::getHW::32] Error Code 2: Internal Error (Assertion d.nbDims >= 2 failed. )
[03/06/2023-09:32:44] [TRT] [E] 3: (Unnamed Layer* 45) [Convolution]:kernel weights has count 1728 but 331776 was expected
[03/06/2023-09:32:44] [TRT] [E] 4: (Unnamed Layer* 45) [Convolution]: count of 1728 weights in kernel, but kernel dimensions (3,3) with 192 input channels, 192 output channels and 1 groups were specified. Expected Weights count is 192 * 3*3 * 192 / 1 = 331776
[03/06/2023-09:32:44] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 45) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:44] [TRT] [E] 3: (Unnamed Layer* 45) [Convolution]:kernel weights has count 1728 but 331776 was expected
[03/06/2023-09:32:44] [TRT] [E] 4: (Unnamed Layer* 45) [Convolution]: count of 1728 weights in kernel, but kernel dimensions (3,3) with 192 input channels, 192 output channels and 1 groups were specified. Expected Weights count is 192 * 3*3 * 192 / 1 = 331776
[03/06/2023-09:32:44] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 45) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:44] [TRT] [E] 3: (Unnamed Layer* 45) [Convolution]:kernel weights has count 1728 but 331776 was expected
[03/06/2023-09:32:44] [TRT] [E] 4: (Unnamed Layer* 45) [Convolution]: count of 1728 weights in kernel, but kernel dimensions (3,3) with 192 input channels, 192 output channels and 1 groups were specified. Expected Weights count is 192 * 3*3 * 192 / 1 = 331776
[03/06/2023-09:32:44] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 45) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:44] [TRT] [E] 2: [helpers.cpp::getHW::32] Error Code 2: Internal Error (Assertion d.nbDims >= 2 failed. )
[03/06/2023-09:32:44] [TRT] [E] 3: (Unnamed Layer* 54) [Convolution]:kernel weights has count 1728 but 331776 was expected
[03/06/2023-09:32:44] [TRT] [E] 4: (Unnamed Layer* 54) [Convolution]: count of 1728 weights in kernel, but kernel dimensions (3,3) with 192 input channels, 192 output channels and 1 groups were specified. Expected Weights count is 192 * 3*3 * 192 / 1 = 331776
[03/06/2023-09:32:45] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 54) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:45] [TRT] [E] 3: (Unnamed Layer* 54) [Convolution]:kernel weights has count 1728 but 331776 was expected
[03/06/2023-09:32:45] [TRT] [E] 4: (Unnamed Layer* 54) [Convolution]: count of 1728 weights in kernel, but kernel dimensions (3,3) with 192 input channels, 192 output channels and 1 groups were specified. Expected Weights count is 192 * 3*3 * 192 / 1 = 331776
[03/06/2023-09:32:45] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 54) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:45] [TRT] [E] 3: (Unnamed Layer* 54) [Convolution]:kernel weights has count 1728 but 331776 was expected
[03/06/2023-09:32:45] [TRT] [E] 4: (Unnamed Layer* 54) [Convolution]: count of 1728 weights in kernel, but kernel dimensions (3,3) with 192 input channels, 192 output channels and 1 groups were specified. Expected Weights count is 192 * 3*3 * 192 / 1 = 331776
[03/06/2023-09:32:45] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 54) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:45] [TRT] [E] 2: [helpers.cpp::getHW::32] Error Code 2: Internal Error (Assertion d.nbDims >= 2 failed. )
[03/06/2023-09:32:45] [TRT] [E] 3: (Unnamed Layer* 62) [Convolution]:kernel weights has count 3456 but 1327104 was expected
[03/06/2023-09:32:45] [TRT] [E] 4: (Unnamed Layer* 62) [Convolution]: count of 3456 weights in kernel, but kernel dimensions (3,3) with 384 input channels, 384 output channels and 1 groups were specified. Expected Weights count is 384 * 3*3 * 384 / 1 = 1327104
[03/06/2023-09:32:45] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 62) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:45] [TRT] [E] 3: (Unnamed Layer* 62) [Convolution]:kernel weights has count 3456 but 1327104 was expected
[03/06/2023-09:32:45] [TRT] [E] 4: (Unnamed Layer* 62) [Convolution]: count of 3456 weights in kernel, but kernel dimensions (3,3) with 384 input channels, 384 output channels and 1 groups were specified. Expected Weights count is 384 * 3*3 * 384 / 1 = 1327104
[03/06/2023-09:32:45] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 62) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:45] [TRT] [E] 3: (Unnamed Layer* 62) [Convolution]:kernel weights has count 3456 but 1327104 was expected
[03/06/2023-09:32:45] [TRT] [E] 4: (Unnamed Layer* 62) [Convolution]: count of 3456 weights in kernel, but kernel dimensions (3,3) with 384 input channels, 384 output channels and 1 groups were specified. Expected Weights count is 384 * 3*3 * 384 / 1 = 1327104
[03/06/2023-09:32:45] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 62) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:45] [TRT] [E] 2: [helpers.cpp::getHW::32] Error Code 2: Internal Error (Assertion d.nbDims >= 2 failed. )
[03/06/2023-09:32:45] [TRT] [E] 3: (Unnamed Layer* 71) [Convolution]:kernel weights has count 3456 but 1327104 was expected
[03/06/2023-09:32:45] [TRT] [E] 4: (Unnamed Layer* 71) [Convolution]: count of 3456 weights in kernel, but kernel dimensions (3,3) with 384 input channels, 384 output channels and 1 groups were specified. Expected Weights count is 384 * 3*3 * 384 / 1 = 1327104
[03/06/2023-09:32:45] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 71) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:45] [TRT] [E] 3: (Unnamed Layer* 71) [Convolution]:kernel weights has count 3456 but 1327104 was expected
[03/06/2023-09:32:45] [TRT] [E] 4: (Unnamed Layer* 71) [Convolution]: count of 3456 weights in kernel, but kernel dimensions (3,3) with 384 input channels, 384 output channels and 1 groups were specified. Expected Weights count is 384 * 3*3 * 384 / 1 = 1327104
[03/06/2023-09:32:46] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 71) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:46] [TRT] [E] 3: (Unnamed Layer* 71) [Convolution]:kernel weights has count 3456 but 1327104 was expected
[03/06/2023-09:32:46] [TRT] [E] 4: (Unnamed Layer* 71) [Convolution]: count of 3456 weights in kernel, but kernel dimensions (3,3) with 384 input channels, 384 output channels and 1 groups were specified. Expected Weights count is 384 * 3*3 * 384 / 1 = 1327104
[03/06/2023-09:32:46] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 71) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:46] [TRT] [E] 2: [helpers.cpp::getHW::32] Error Code 2: Internal Error (Assertion d.nbDims >= 2 failed. )
[03/06/2023-09:32:46] [TRT] [E] 3: (Unnamed Layer* 80) [Convolution]:kernel weights has count 3456 but 1327104 was expected
[03/06/2023-09:32:46] [TRT] [E] 4: (Unnamed Layer* 80) [Convolution]: count of 3456 weights in kernel, but kernel dimensions (3,3) with 384 input channels, 384 output channels and 1 groups were specified. Expected Weights count is 384 * 3*3 * 384 / 1 = 1327104
[03/06/2023-09:32:46] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 80) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:46] [TRT] [E] 3: (Unnamed Layer* 80) [Convolution]:kernel weights has count 3456 but 1327104 was expected
[03/06/2023-09:32:46] [TRT] [E] 4: (Unnamed Layer* 80) [Convolution]: count of 3456 weights in kernel, but kernel dimensions (3,3) with 384 input channels, 384 output channels and 1 groups were specified. Expected Weights count is 384 * 3*3 * 384 / 1 = 1327104
[03/06/2023-09:32:46] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 80) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:46] [TRT] [E] 3: (Unnamed Layer* 80) [Convolution]:kernel weights has count 3456 but 1327104 was expected
[03/06/2023-09:32:46] [TRT] [E] 4: (Unnamed Layer* 80) [Convolution]: count of 3456 weights in kernel, but kernel dimensions (3,3) with 384 input channels, 384 output channels and 1 groups were specified. Expected Weights count is 384 * 3*3 * 384 / 1 = 1327104
[03/06/2023-09:32:46] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 80) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:46] [TRT] [E] 2: [helpers.cpp::getHW::32] Error Code 2: Internal Error (Assertion d.nbDims >= 2 failed. )
[03/06/2023-09:32:46] [TRT] [E] 3: (Unnamed Layer* 89) [Convolution]:kernel weights has count 3456 but 1327104 was expected
[03/06/2023-09:32:46] [TRT] [E] 4: (Unnamed Layer* 89) [Convolution]: count of 3456 weights in kernel, but kernel dimensions (3,3) with 384 input channels, 384 output channels and 1 groups were specified. Expected Weights count is 384 * 3*3 * 384 / 1 = 1327104
[03/06/2023-09:32:46] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 89) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:46] [TRT] [E] 3: (Unnamed Layer* 89) [Convolution]:kernel weights has count 3456 but 1327104 was expected
[03/06/2023-09:32:46] [TRT] [E] 4: (Unnamed Layer* 89) [Convolution]: count of 3456 weights in kernel, but kernel dimensions (3,3) with 384 input channels, 384 output channels and 1 groups were specified. Expected Weights count is 384 * 3*3 * 384 / 1 = 1327104
[03/06/2023-09:32:46] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 89) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:46] [TRT] [E] 3: (Unnamed Layer* 89) [Convolution]:kernel weights has count 3456 but 1327104 was expected
[03/06/2023-09:32:46] [TRT] [E] 4: (Unnamed Layer* 89) [Convolution]: count of 3456 weights in kernel, but kernel dimensions (3,3) with 384 input channels, 384 output channels and 1 groups were specified. Expected Weights count is 384 * 3*3 * 384 / 1 = 1327104
[03/06/2023-09:32:46] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 89) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:46] [TRT] [E] 2: [helpers.cpp::getHW::32] Error Code 2: Internal Error (Assertion d.nbDims >= 2 failed. )
[03/06/2023-09:32:46] [TRT] [E] 3: (Unnamed Layer* 97) [Convolution]:kernel weights has count 5184 but 2985984 was expected
[03/06/2023-09:32:46] [TRT] [E] 4: (Unnamed Layer* 97) [Convolution]: count of 5184 weights in kernel, but kernel dimensions (3,3) with 576 input channels, 576 output channels and 1 groups were specified. Expected Weights count is 576 * 3*3 * 576 / 1 = 2985984
[03/06/2023-09:32:47] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 97) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:47] [TRT] [E] 3: (Unnamed Layer* 97) [Convolution]:kernel weights has count 5184 but 2985984 was expected
[03/06/2023-09:32:47] [TRT] [E] 4: (Unnamed Layer* 97) [Convolution]: count of 5184 weights in kernel, but kernel dimensions (3,3) with 576 input channels, 576 output channels and 1 groups were specified. Expected Weights count is 576 * 3*3 * 576 / 1 = 2985984
[03/06/2023-09:32:47] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 97) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:47] [TRT] [E] 3: (Unnamed Layer* 97) [Convolution]:kernel weights has count 5184 but 2985984 was expected
[03/06/2023-09:32:47] [TRT] [E] 4: (Unnamed Layer* 97) [Convolution]: count of 5184 weights in kernel, but kernel dimensions (3,3) with 576 input channels, 576 output channels and 1 groups were specified. Expected Weights count is 576 * 3*3 * 576 / 1 = 2985984
[03/06/2023-09:32:47] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 97) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:47] [TRT] [E] 2: [helpers.cpp::getHW::32] Error Code 2: Internal Error (Assertion d.nbDims >= 2 failed. )
[03/06/2023-09:32:47] [TRT] [E] 3: (Unnamed Layer* 106) [Convolution]:kernel weights has count 5184 but 2985984 was expected
[03/06/2023-09:32:47] [TRT] [E] 4: (Unnamed Layer* 106) [Convolution]: count of 5184 weights in kernel, but kernel dimensions (3,3) with 576 input channels, 576 output channels and 1 groups were specified. Expected Weights count is 576 * 3*3 * 576 / 1 = 2985984
[03/06/2023-09:32:47] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 106) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:47] [TRT] [E] 3: (Unnamed Layer* 106) [Convolution]:kernel weights has count 5184 but 2985984 was expected
[03/06/2023-09:32:47] [TRT] [E] 4: (Unnamed Layer* 106) [Convolution]: count of 5184 weights in kernel, but kernel dimensions (3,3) with 576 input channels, 576 output channels and 1 groups were specified. Expected Weights count is 576 * 3*3 * 576 / 1 = 2985984
[03/06/2023-09:32:47] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 106) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:47] [TRT] [E] 3: (Unnamed Layer* 106) [Convolution]:kernel weights has count 5184 but 2985984 was expected
[03/06/2023-09:32:47] [TRT] [E] 4: (Unnamed Layer* 106) [Convolution]: count of 5184 weights in kernel, but kernel dimensions (3,3) with 576 input channels, 576 output channels and 1 groups were specified. Expected Weights count is 576 * 3*3 * 576 / 1 = 2985984
[03/06/2023-09:32:47] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 106) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:47] [TRT] [E] 2: [helpers.cpp::getHW::32] Error Code 2: Internal Error (Assertion d.nbDims >= 2 failed. )
[03/06/2023-09:32:48] [TRT] [E] 3: (Unnamed Layer* 117) [Convolution]:kernel weights has count 171 but 3249 was expected
[03/06/2023-09:32:48] [TRT] [E] 4: (Unnamed Layer* 117) [Convolution]: count of 171 weights in kernel, but kernel dimensions (3,3) with 19 input channels, 19 output channels and 1 groups were specified. Expected Weights count is 19 * 3*3 * 19 / 1 = 3249
[03/06/2023-09:32:48] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 117) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:48] [TRT] [E] 3: (Unnamed Layer* 117) [Convolution]:kernel weights has count 171 but 3249 was expected
[03/06/2023-09:32:48] [TRT] [E] 4: (Unnamed Layer* 117) [Convolution]: count of 171 weights in kernel, but kernel dimensions (3,3) with 19 input channels, 19 output channels and 1 groups were specified. Expected Weights count is 19 * 3*3 * 19 / 1 = 3249
[03/06/2023-09:32:48] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 117) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:48] [TRT] [E] 3: (Unnamed Layer* 117) [Convolution]:kernel weights has count 171 but 3249 was expected
[03/06/2023-09:32:48] [TRT] [E] 4: (Unnamed Layer* 117) [Convolution]: count of 171 weights in kernel, but kernel dimensions (3,3) with 19 input channels, 19 output channels and 1 groups were specified. Expected Weights count is 19 * 3*3 * 19 / 1 = 3249
[03/06/2023-09:32:48] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 117) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:48] [TRT] [E] 2: [helpers.cpp::getHW::32] Error Code 2: Internal Error (Assertion d.nbDims >= 2 failed. )
[03/06/2023-09:32:49] [TRT] [E] 3: (Unnamed Layer* 127) [Convolution]:kernel weights has count 117 but 1521 was expected
[03/06/2023-09:32:49] [TRT] [E] 4: (Unnamed Layer* 127) [Convolution]: count of 117 weights in kernel, but kernel dimensions (3,3) with 13 input channels, 13 output channels and 1 groups were specified. Expected Weights count is 13 * 3*3 * 13 / 1 = 1521
[03/06/2023-09:32:49] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 127) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:49] [TRT] [E] 3: (Unnamed Layer* 127) [Convolution]:kernel weights has count 117 but 1521 was expected
[03/06/2023-09:32:49] [TRT] [E] 4: (Unnamed Layer* 127) [Convolution]: count of 117 weights in kernel, but kernel dimensions (3,3) with 13 input channels, 13 output channels and 1 groups were specified. Expected Weights count is 13 * 3*3 * 13 / 1 = 1521
[03/06/2023-09:32:49] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 127) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:49] [TRT] [E] 3: (Unnamed Layer* 127) [Convolution]:kernel weights has count 117 but 1521 was expected
[03/06/2023-09:32:49] [TRT] [E] 4: (Unnamed Layer* 127) [Convolution]: count of 117 weights in kernel, but kernel dimensions (3,3) with 13 input channels, 13 output channels and 1 groups were specified. Expected Weights count is 13 * 3*3 * 13 / 1 = 1521
[03/06/2023-09:32:49] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 127) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:49] [TRT] [E] 2: [helpers.cpp::getHW::32] Error Code 2: Internal Error (Assertion d.nbDims >= 2 failed. )
[03/06/2023-09:32:50] [TRT] [E] 3: (Unnamed Layer* 137) [Convolution]:kernel weights has count 81 but 729 was expected
[03/06/2023-09:32:50] [TRT] [E] 4: (Unnamed Layer* 137) [Convolution]: count of 81 weights in kernel, but kernel dimensions (3,3) with 9 input channels, 9 output channels and 1 groups were specified. Expected Weights count is 9 * 3*3 * 9 / 1 = 729
[03/06/2023-09:32:50] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 137) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:50] [TRT] [E] 3: (Unnamed Layer* 137) [Convolution]:kernel weights has count 81 but 729 was expected
[03/06/2023-09:32:50] [TRT] [E] 4: (Unnamed Layer* 137) [Convolution]: count of 81 weights in kernel, but kernel dimensions (3,3) with 9 input channels, 9 output channels and 1 groups were specified. Expected Weights count is 9 * 3*3 * 9 / 1 = 729
[03/06/2023-09:32:50] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 137) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:50] [TRT] [E] 3: (Unnamed Layer* 137) [Convolution]:kernel weights has count 81 but 729 was expected
[03/06/2023-09:32:50] [TRT] [E] 4: (Unnamed Layer* 137) [Convolution]: count of 81 weights in kernel, but kernel dimensions (3,3) with 9 input channels, 9 output channels and 1 groups were specified. Expected Weights count is 9 * 3*3 * 9 / 1 = 729
[03/06/2023-09:32:50] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 137) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:50] [TRT] [E] 2: [helpers.cpp::getHW::32] Error Code 2: Internal Error (Assertion d.nbDims >= 2 failed. )
[03/06/2023-09:32:52] [TRT] [E] 3: (Unnamed Layer* 147) [Convolution]:kernel weights has count 45 but 225 was expected
[03/06/2023-09:32:52] [TRT] [E] 4: (Unnamed Layer* 147) [Convolution]: count of 45 weights in kernel, but kernel dimensions (3,3) with 5 input channels, 5 output channels and 1 groups were specified. Expected Weights count is 5 * 3*3 * 5 / 1 = 225
[03/06/2023-09:32:52] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 147) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:52] [TRT] [E] 3: (Unnamed Layer* 147) [Convolution]:kernel weights has count 45 but 225 was expected
[03/06/2023-09:32:52] [TRT] [E] 4: (Unnamed Layer* 147) [Convolution]: count of 45 weights in kernel, but kernel dimensions (3,3) with 5 input channels, 5 output channels and 1 groups were specified. Expected Weights count is 5 * 3*3 * 5 / 1 = 225
[03/06/2023-09:32:52] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 147) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:52] [TRT] [E] 3: (Unnamed Layer* 147) [Convolution]:kernel weights has count 45 but 225 was expected
[03/06/2023-09:32:52] [TRT] [E] 4: (Unnamed Layer* 147) [Convolution]: count of 45 weights in kernel, but kernel dimensions (3,3) with 5 input channels, 5 output channels and 1 groups were specified. Expected Weights count is 5 * 3*3 * 5 / 1 = 225
[03/06/2023-09:32:52] [TRT] [E] 4: [convolutionNode.cpp::computeOutputExtents::42] Error Code 4: Internal Error ((Unnamed Layer* 147) [Convolution]: number of kernel weights does not match tensor dimensions)
[03/06/2023-09:32:52] [TRT] [E] 2: [helpers.cpp::getHW::32] Error Code 2: Internal Error (Assertion d.nbDims >= 2 failed. )

The correct .pth file is this one.

Thanks!

AastaLLL · March 16, 2023, 3:17am

Hi,

Based on the below log, it looks like TensorRT expects the kernel number to be 32x32 but the real number is 1x32.

[03/06/2023-09:32:42] [TRT] [E] 3: (Unnamed Layer* 3) [Convolution]:kernel weights has count 288 but 9216 was expected
[03/06/2023-09:32:42] [TRT] [E] 4: (Unnamed Layer* 3) [Convolution]: count of 288 weights in kernel, but kernel dimensions (3,3) with 32 input channels, 32 output channels and 1 groups were specified. Expected Weights count is 32 * 3*3 * 32 / 1 = 9216

Is this issue can be reproduced with the ONNX model below:

Thanks.

kmarx · March 16, 2023, 10:57pm

I have tried using torch.onnx.export to convert this model from torch to onnx and converting from onnx to TensorRT using trtexec

My code for converting from torch to onnx is the following:

import torch
from m2unet import M2UNet, Encoder

def main():
    modelpath = "models/flat_model_56_10.pth"
    savepath = "models/flat_model_56_10.onnx"
    NCH = 1
    BATCHSZ = 1
    im_sz = 1024
    v = 12
    sz = (BATCHSZ, NCH, im_sz, im_sz) # size of input - batch, channels, x, y

    # define your PyTorch model and load the weights
    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
    encoder = Encoder(NUM_CHAN=NCH)
    model = M2UNet(encoder, NUM_CHAN=NCH).float().to(device).cuda().eval()
    model.load_state_dict(torch.load(modelpath))

    # create an example input
    input_data = torch.ones(sz).cuda()
    
    # export
    torch.onnx.export(model, input_data, savepath, verbose=True, opset_version=v)
    

if __name__ == "__main__":
    main()

Running this produced the following logs:

/home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/onnx/_internal/jit_utils.py:258: UserWarning: The shape inference of prim::Constant type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function. (Triggered internally at ../torch/csrc/jit/passes/onnx/shape_type_inference.cpp:1884.)
  _C._jit_pass_onnx_node_shape_type_inference(node, params_dict, opset_version)
/home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/onnx/utils.py:687: UserWarning: The shape inference of prim::Constant type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function. (Triggered internally at ../torch/csrc/jit/passes/onnx/shape_type_inference.cpp:1884.)
  _C._jit_pass_onnx_graph_shape_type_inference(
/home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/onnx/utils.py:1178: UserWarning: The shape inference of prim::Constant type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function. (Triggered internally at ../torch/csrc/jit/passes/onnx/shape_type_inference.cpp:1884.)
  _C._jit_pass_onnx_graph_shape_type_inference(
Exported graph: graph(%input.1 : Float(1, 1, 1024, 1024, strides=[1048576, 1048576, 1024, 1], requires_grad=0, device=cuda:0),
      %decode1.conv.0.weight : Float(1, 16, 1, 1, strides=[16, 1, 1, 1], requires_grad=1, device=cuda:0),
      %decode1.conv.0.bias : Float(1, strides=[1], requires_grad=1, device=cuda:0),
      %onnx::Conv_479 : Float(32, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_480 : Float(32, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_482 : Float(32, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_483 : Float(32, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_485 : Float(16, 32, 1, 1, strides=[32, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_486 : Float(16, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_488 : Float(96, 16, 1, 1, strides=[16, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_489 : Float(96, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_491 : Float(96, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_492 : Float(96, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_494 : Float(24, 96, 1, 1, strides=[96, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_495 : Float(24, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_497 : Float(144, 24, 1, 1, strides=[24, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_498 : Float(144, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_500 : Float(144, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_501 : Float(144, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_503 : Float(24, 144, 1, 1, strides=[144, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_504 : Float(24, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_506 : Float(144, 24, 1, 1, strides=[24, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_507 : Float(144, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_509 : Float(144, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_510 : Float(144, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_512 : Float(32, 144, 1, 1, strides=[144, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_513 : Float(32, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_515 : Float(192, 32, 1, 1, strides=[32, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_516 : Float(192, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_518 : Float(192, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_519 : Float(192, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_521 : Float(32, 192, 1, 1, strides=[192, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_522 : Float(32, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_524 : Float(192, 32, 1, 1, strides=[32, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_525 : Float(192, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_527 : Float(192, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_528 : Float(192, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_530 : Float(32, 192, 1, 1, strides=[192, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_531 : Float(32, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_533 : Float(192, 32, 1, 1, strides=[32, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_534 : Float(192, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_536 : Float(192, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_537 : Float(192, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_539 : Float(64, 192, 1, 1, strides=[192, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_540 : Float(64, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_542 : Float(384, 64, 1, 1, strides=[64, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_543 : Float(384, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_545 : Float(384, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_546 : Float(384, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_548 : Float(64, 384, 1, 1, strides=[384, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_549 : Float(64, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_551 : Float(384, 64, 1, 1, strides=[64, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_552 : Float(384, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_554 : Float(384, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_555 : Float(384, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_557 : Float(64, 384, 1, 1, strides=[384, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_558 : Float(64, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_560 : Float(384, 64, 1, 1, strides=[64, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_561 : Float(384, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_563 : Float(384, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_564 : Float(384, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_566 : Float(64, 384, 1, 1, strides=[384, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_567 : Float(64, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_569 : Float(384, 64, 1, 1, strides=[64, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_570 : Float(384, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_572 : Float(384, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_573 : Float(384, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_575 : Float(96, 384, 1, 1, strides=[384, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_576 : Float(96, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_578 : Float(576, 96, 1, 1, strides=[96, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_579 : Float(576, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_581 : Float(576, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_582 : Float(576, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_584 : Float(96, 576, 1, 1, strides=[576, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_585 : Float(96, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_587 : Float(576, 96, 1, 1, strides=[96, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_588 : Float(576, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_590 : Float(576, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_591 : Float(576, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_593 : Float(96, 576, 1, 1, strides=[576, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_594 : Float(96, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_596 : Float(19, 128, 1, 1, strides=[128, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_597 : Float(19, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_599 : Float(19, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_600 : Float(19, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_602 : Float(64, 19, 1, 1, strides=[19, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_603 : Float(64, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_605 : Float(13, 88, 1, 1, strides=[88, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_606 : Float(13, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_608 : Float(13, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_609 : Float(13, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_611 : Float(44, 13, 1, 1, strides=[13, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_612 : Float(44, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_614 : Float(9, 60, 1, 1, strides=[60, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_615 : Float(9, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_617 : Float(9, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_618 : Float(9, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_620 : Float(30, 9, 1, 1, strides=[9, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_621 : Float(30, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_623 : Float(5, 31, 1, 1, strides=[31, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_624 : Float(5, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_626 : Float(5, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_627 : Float(5, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_629 : Float(16, 5, 1, 1, strides=[5, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_630 : Float(16, strides=[1], requires_grad=0, device=cuda:0)):
  %/conv1/conv1.0/conv1.0.0/Conv_output_0 : Float(1, 32, 512, 512, strides=[8388608, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[2, 2], onnx_name="/conv1/conv1.0/conv1.0.0/Conv"](%input.1, %onnx::Conv_479, %onnx::Conv_480), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv1/torch.nn.modules.container.Sequential::conv1.0/torch.nn.modules.conv.Conv2d::conv1.0.0 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv1/conv1.0/conv1.0.2/Relu_output_0 : Float(1, 32, 512, 512, strides=[8388608, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv1/conv1.0/conv1.0.2/Relu"](%/conv1/conv1.0/conv1.0.0/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv1/torch.nn.modules.container.Sequential::conv1.0/torch.nn.modules.activation.ReLU::conv1.0.2 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv1/conv1.1/conv/conv.0/Conv_output_0 : Float(1, 32, 512, 512, strides=[8388608, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=32, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/conv1/conv1.1/conv/conv.0/Conv"](%/conv1/conv1.0/conv1.0.2/Relu_output_0, %onnx::Conv_482, %onnx::Conv_483), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv1/m2unet.InvertedResidual::conv1.1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv1/conv1.1/conv/conv.2/Relu_output_0 : Float(1, 32, 512, 512, strides=[8388608, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv1/conv1.1/conv/conv.2/Relu"](%/conv1/conv1.1/conv/conv.0/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv1/m2unet.InvertedResidual::conv1.1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.2 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv1/conv1.1/conv/conv.3/Conv_output_0 : Float(1, 16, 512, 512, strides=[4194304, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv1/conv1.1/conv/conv.3/Conv"](%/conv1/conv1.1/conv/conv.2/Relu_output_0, %onnx::Conv_485, %onnx::Conv_486), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv1/m2unet.InvertedResidual::conv1.1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv2/conv2.2/conv/conv.0/Conv_output_0 : Float(1, 96, 512, 512, strides=[25165824, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv2/conv2.2/conv/conv.0/Conv"](%/conv1/conv1.1/conv/conv.3/Conv_output_0, %onnx::Conv_488, %onnx::Conv_489), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet.InvertedResidual::conv2.2/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv2/conv2.2/conv/conv.2/Relu_output_0 : Float(1, 96, 512, 512, strides=[25165824, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv2/conv2.2/conv/conv.2/Relu"](%/conv2/conv2.2/conv/conv.0/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet.InvertedResidual::conv2.2/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.2 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv2/conv2.2/conv/conv.3/Conv_output_0 : Float(1, 96, 256, 256, strides=[6291456, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=96, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[2, 2], onnx_name="/conv2/conv2.2/conv/conv.3/Conv"](%/conv2/conv2.2/conv/conv.2/Relu_output_0, %onnx::Conv_491, %onnx::Conv_492), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet.InvertedResidual::conv2.2/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv2/conv2.2/conv/conv.5/Relu_output_0 : Float(1, 96, 256, 256, strides=[6291456, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv2/conv2.2/conv/conv.5/Relu"](%/conv2/conv2.2/conv/conv.3/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet.InvertedResidual::conv2.2/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.5 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv2/conv2.2/conv/conv.6/Conv_output_0 : Float(1, 24, 256, 256, strides=[1572864, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv2/conv2.2/conv/conv.6/Conv"](%/conv2/conv2.2/conv/conv.5/Relu_output_0, %onnx::Conv_494, %onnx::Conv_495), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet.InvertedResidual::conv2.2/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv2/conv2.3/conv/conv.0/Conv_output_0 : Float(1, 144, 256, 256, strides=[9437184, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv2/conv2.3/conv/conv.0/Conv"](%/conv2/conv2.2/conv/conv.6/Conv_output_0, %onnx::Conv_497, %onnx::Conv_498), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet.InvertedResidual::conv2.3/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv2/conv2.3/conv/conv.2/Relu_output_0 : Float(1, 144, 256, 256, strides=[9437184, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv2/conv2.3/conv/conv.2/Relu"](%/conv2/conv2.3/conv/conv.0/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet.InvertedResidual::conv2.3/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.2 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv2/conv2.3/conv/conv.3/Conv_output_0 : Float(1, 144, 256, 256, strides=[9437184, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=144, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/conv2/conv2.3/conv/conv.3/Conv"](%/conv2/conv2.3/conv/conv.2/Relu_output_0, %onnx::Conv_500, %onnx::Conv_501), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet.InvertedResidual::conv2.3/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv2/conv2.3/conv/conv.5/Relu_output_0 : Float(1, 144, 256, 256, strides=[9437184, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv2/conv2.3/conv/conv.5/Relu"](%/conv2/conv2.3/conv/conv.3/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet.InvertedResidual::conv2.3/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.5 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv2/conv2.3/conv/conv.6/Conv_output_0 : Float(1, 24, 256, 256, strides=[1572864, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv2/conv2.3/conv/conv.6/Conv"](%/conv2/conv2.3/conv/conv.5/Relu_output_0, %onnx::Conv_503, %onnx::Conv_504), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet.InvertedResidual::conv2.3/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv2/conv2.3/Add_output_0 : Float(1, 24, 256, 256, strides=[1572864, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Add[onnx_name="/conv2/conv2.3/Add"](%/conv2/conv2.2/conv/conv.6/Conv_output_0, %/conv2/conv2.3/conv/conv.6/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet.InvertedResidual::conv2.3 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/ao/nn/quantized/modules/functional_modules.py:45:0
  %/conv3/conv3.4/conv/conv.0/Conv_output_0 : Float(1, 144, 256, 256, strides=[9437184, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv3/conv3.4/conv/conv.0/Conv"](%/conv2/conv2.3/Add_output_0, %onnx::Conv_506, %onnx::Conv_507), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet.InvertedResidual::conv3.4/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv3/conv3.4/conv/conv.2/Relu_output_0 : Float(1, 144, 256, 256, strides=[9437184, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv3/conv3.4/conv/conv.2/Relu"](%/conv3/conv3.4/conv/conv.0/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet.InvertedResidual::conv3.4/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.2 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv3/conv3.4/conv/conv.3/Conv_output_0 : Float(1, 144, 128, 128, strides=[2359296, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=144, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[2, 2], onnx_name="/conv3/conv3.4/conv/conv.3/Conv"](%/conv3/conv3.4/conv/conv.2/Relu_output_0, %onnx::Conv_509, %onnx::Conv_510), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet.InvertedResidual::conv3.4/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv3/conv3.4/conv/conv.5/Relu_output_0 : Float(1, 144, 128, 128, strides=[2359296, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv3/conv3.4/conv/conv.5/Relu"](%/conv3/conv3.4/conv/conv.3/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet.InvertedResidual::conv3.4/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.5 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv3/conv3.4/conv/conv.6/Conv_output_0 : Float(1, 32, 128, 128, strides=[524288, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv3/conv3.4/conv/conv.6/Conv"](%/conv3/conv3.4/conv/conv.5/Relu_output_0, %onnx::Conv_512, %onnx::Conv_513), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet.InvertedResidual::conv3.4/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv3/conv3.5/conv/conv.0/Conv_output_0 : Float(1, 192, 128, 128, strides=[3145728, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv3/conv3.5/conv/conv.0/Conv"](%/conv3/conv3.4/conv/conv.6/Conv_output_0, %onnx::Conv_515, %onnx::Conv_516), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet.InvertedResidual::conv3.5/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv3/conv3.5/conv/conv.2/Relu_output_0 : Float(1, 192, 128, 128, strides=[3145728, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv3/conv3.5/conv/conv.2/Relu"](%/conv3/conv3.5/conv/conv.0/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet.InvertedResidual::conv3.5/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.2 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv3/conv3.5/conv/conv.3/Conv_output_0 : Float(1, 192, 128, 128, strides=[3145728, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=192, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/conv3/conv3.5/conv/conv.3/Conv"](%/conv3/conv3.5/conv/conv.2/Relu_output_0, %onnx::Conv_518, %onnx::Conv_519), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet.InvertedResidual::conv3.5/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv3/conv3.5/conv/conv.5/Relu_output_0 : Float(1, 192, 128, 128, strides=[3145728, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv3/conv3.5/conv/conv.5/Relu"](%/conv3/conv3.5/conv/conv.3/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet.InvertedResidual::conv3.5/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.5 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv3/conv3.5/conv/conv.6/Conv_output_0 : Float(1, 32, 128, 128, strides=[524288, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv3/conv3.5/conv/conv.6/Conv"](%/conv3/conv3.5/conv/conv.5/Relu_output_0, %onnx::Conv_521, %onnx::Conv_522), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet.InvertedResidual::conv3.5/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv3/conv3.5/Add_output_0 : Float(1, 32, 128, 128, strides=[524288, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Add[onnx_name="/conv3/conv3.5/Add"](%/conv3/conv3.4/conv/conv.6/Conv_output_0, %/conv3/conv3.5/conv/conv.6/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet.InvertedResidual::conv3.5 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/ao/nn/quantized/modules/functional_modules.py:45:0
  %/conv3/conv3.6/conv/conv.0/Conv_output_0 : Float(1, 192, 128, 128, strides=[3145728, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv3/conv3.6/conv/conv.0/Conv"](%/conv3/conv3.5/Add_output_0, %onnx::Conv_524, %onnx::Conv_525), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet.InvertedResidual::conv3.6/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv3/conv3.6/conv/conv.2/Relu_output_0 : Float(1, 192, 128, 128, strides=[3145728, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv3/conv3.6/conv/conv.2/Relu"](%/conv3/conv3.6/conv/conv.0/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet.InvertedResidual::conv3.6/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.2 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv3/conv3.6/conv/conv.3/Conv_output_0 : Float(1, 192, 128, 128, strides=[3145728, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=192, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/conv3/conv3.6/conv/conv.3/Conv"](%/conv3/conv3.6/conv/conv.2/Relu_output_0, %onnx::Conv_527, %onnx::Conv_528), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet.InvertedResidual::conv3.6/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv3/conv3.6/conv/conv.5/Relu_output_0 : Float(1, 192, 128, 128, strides=[3145728, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv3/conv3.6/conv/conv.5/Relu"](%/conv3/conv3.6/conv/conv.3/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet.InvertedResidual::conv3.6/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.5 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv3/conv3.6/conv/conv.6/Conv_output_0 : Float(1, 32, 128, 128, strides=[524288, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv3/conv3.6/conv/conv.6/Conv"](%/conv3/conv3.6/conv/conv.5/Relu_output_0, %onnx::Conv_530, %onnx::Conv_531), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet.InvertedResidual::conv3.6/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv3/conv3.6/Add_output_0 : Float(1, 32, 128, 128, strides=[524288, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Add[onnx_name="/conv3/conv3.6/Add"](%/conv3/conv3.5/Add_output_0, %/conv3/conv3.6/conv/conv.6/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet.InvertedResidual::conv3.6 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/ao/nn/quantized/modules/functional_modules.py:45:0
  %/conv4/conv4.7/conv/conv.0/Conv_output_0 : Float(1, 192, 128, 128, strides=[3145728, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv4/conv4.7/conv/conv.0/Conv"](%/conv3/conv3.6/Add_output_0, %onnx::Conv_533, %onnx::Conv_534), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.7/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.7/conv/conv.2/Relu_output_0 : Float(1, 192, 128, 128, strides=[3145728, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv4/conv4.7/conv/conv.2/Relu"](%/conv4/conv4.7/conv/conv.0/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.7/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.2 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv4/conv4.7/conv/conv.3/Conv_output_0 : Float(1, 192, 64, 64, strides=[786432, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=192, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[2, 2], onnx_name="/conv4/conv4.7/conv/conv.3/Conv"](%/conv4/conv4.7/conv/conv.2/Relu_output_0, %onnx::Conv_536, %onnx::Conv_537), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.7/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.7/conv/conv.5/Relu_output_0 : Float(1, 192, 64, 64, strides=[786432, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv4/conv4.7/conv/conv.5/Relu"](%/conv4/conv4.7/conv/conv.3/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.7/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.5 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv4/conv4.7/conv/conv.6/Conv_output_0 : Float(1, 64, 64, 64, strides=[262144, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv4/conv4.7/conv/conv.6/Conv"](%/conv4/conv4.7/conv/conv.5/Relu_output_0, %onnx::Conv_539, %onnx::Conv_540), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.7/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.8/conv/conv.0/Conv_output_0 : Float(1, 384, 64, 64, strides=[1572864, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv4/conv4.8/conv/conv.0/Conv"](%/conv4/conv4.7/conv/conv.6/Conv_output_0, %onnx::Conv_542, %onnx::Conv_543), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.8/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.8/conv/conv.2/Relu_output_0 : Float(1, 384, 64, 64, strides=[1572864, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv4/conv4.8/conv/conv.2/Relu"](%/conv4/conv4.8/conv/conv.0/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.8/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.2 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv4/conv4.8/conv/conv.3/Conv_output_0 : Float(1, 384, 64, 64, strides=[1572864, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=384, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/conv4/conv4.8/conv/conv.3/Conv"](%/conv4/conv4.8/conv/conv.2/Relu_output_0, %onnx::Conv_545, %onnx::Conv_546), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.8/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.8/conv/conv.5/Relu_output_0 : Float(1, 384, 64, 64, strides=[1572864, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv4/conv4.8/conv/conv.5/Relu"](%/conv4/conv4.8/conv/conv.3/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.8/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.5 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv4/conv4.8/conv/conv.6/Conv_output_0 : Float(1, 64, 64, 64, strides=[262144, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv4/conv4.8/conv/conv.6/Conv"](%/conv4/conv4.8/conv/conv.5/Relu_output_0, %onnx::Conv_548, %onnx::Conv_549), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.8/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.8/Add_output_0 : Float(1, 64, 64, 64, strides=[262144, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Add[onnx_name="/conv4/conv4.8/Add"](%/conv4/conv4.7/conv/conv.6/Conv_output_0, %/conv4/conv4.8/conv/conv.6/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.8 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/ao/nn/quantized/modules/functional_modules.py:45:0
  %/conv4/conv4.9/conv/conv.0/Conv_output_0 : Float(1, 384, 64, 64, strides=[1572864, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv4/conv4.9/conv/conv.0/Conv"](%/conv4/conv4.8/Add_output_0, %onnx::Conv_551, %onnx::Conv_552), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.9/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.9/conv/conv.2/Relu_output_0 : Float(1, 384, 64, 64, strides=[1572864, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv4/conv4.9/conv/conv.2/Relu"](%/conv4/conv4.9/conv/conv.0/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.9/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.2 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv4/conv4.9/conv/conv.3/Conv_output_0 : Float(1, 384, 64, 64, strides=[1572864, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=384, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/conv4/conv4.9/conv/conv.3/Conv"](%/conv4/conv4.9/conv/conv.2/Relu_output_0, %onnx::Conv_554, %onnx::Conv_555), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.9/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.9/conv/conv.5/Relu_output_0 : Float(1, 384, 64, 64, strides=[1572864, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv4/conv4.9/conv/conv.5/Relu"](%/conv4/conv4.9/conv/conv.3/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.9/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.5 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv4/conv4.9/conv/conv.6/Conv_output_0 : Float(1, 64, 64, 64, strides=[262144, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv4/conv4.9/conv/conv.6/Conv"](%/conv4/conv4.9/conv/conv.5/Relu_output_0, %onnx::Conv_557, %onnx::Conv_558), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.9/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.9/Add_output_0 : Float(1, 64, 64, 64, strides=[262144, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Add[onnx_name="/conv4/conv4.9/Add"](%/conv4/conv4.8/Add_output_0, %/conv4/conv4.9/conv/conv.6/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.9 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/ao/nn/quantized/modules/functional_modules.py:45:0
  %/conv4/conv4.10/conv/conv.0/Conv_output_0 : Float(1, 384, 64, 64, strides=[1572864, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv4/conv4.10/conv/conv.0/Conv"](%/conv4/conv4.9/Add_output_0, %onnx::Conv_560, %onnx::Conv_561), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.10/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.10/conv/conv.2/Relu_output_0 : Float(1, 384, 64, 64, strides=[1572864, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv4/conv4.10/conv/conv.2/Relu"](%/conv4/conv4.10/conv/conv.0/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.10/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.2 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv4/conv4.10/conv/conv.3/Conv_output_0 : Float(1, 384, 64, 64, strides=[1572864, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=384, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/conv4/conv4.10/conv/conv.3/Conv"](%/conv4/conv4.10/conv/conv.2/Relu_output_0, %onnx::Conv_563, %onnx::Conv_564), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.10/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.10/conv/conv.5/Relu_output_0 : Float(1, 384, 64, 64, strides=[1572864, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv4/conv4.10/conv/conv.5/Relu"](%/conv4/conv4.10/conv/conv.3/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.10/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.5 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv4/conv4.10/conv/conv.6/Conv_output_0 : Float(1, 64, 64, 64, strides=[262144, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv4/conv4.10/conv/conv.6/Conv"](%/conv4/conv4.10/conv/conv.5/Relu_output_0, %onnx::Conv_566, %onnx::Conv_567), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.10/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.10/Add_output_0 : Float(1, 64, 64, 64, strides=[262144, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Add[onnx_name="/conv4/conv4.10/Add"](%/conv4/conv4.9/Add_output_0, %/conv4/conv4.10/conv/conv.6/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.10 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/ao/nn/quantized/modules/functional_modules.py:45:0
  %/conv4/conv4.11/conv/conv.0/Conv_output_0 : Float(1, 384, 64, 64, strides=[1572864, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv4/conv4.11/conv/conv.0/Conv"](%/conv4/conv4.10/Add_output_0, %onnx::Conv_569, %onnx::Conv_570), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.11/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.11/conv/conv.2/Relu_output_0 : Float(1, 384, 64, 64, strides=[1572864, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv4/conv4.11/conv/conv.2/Relu"](%/conv4/conv4.11/conv/conv.0/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.11/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.2 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv4/conv4.11/conv/conv.3/Conv_output_0 : Float(1, 384, 64, 64, strides=[1572864, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=384, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/conv4/conv4.11/conv/conv.3/Conv"](%/conv4/conv4.11/conv/conv.2/Relu_output_0, %onnx::Conv_572, %onnx::Conv_573), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.11/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.11/conv/conv.5/Relu_output_0 : Float(1, 384, 64, 64, strides=[1572864, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv4/conv4.11/conv/conv.5/Relu"](%/conv4/conv4.11/conv/conv.3/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.11/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.5 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv4/conv4.11/conv/conv.6/Conv_output_0 : Float(1, 96, 64, 64, strides=[393216, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv4/conv4.11/conv/conv.6/Conv"](%/conv4/conv4.11/conv/conv.5/Relu_output_0, %onnx::Conv_575, %onnx::Conv_576), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.11/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.12/conv/conv.0/Conv_output_0 : Float(1, 576, 64, 64, strides=[2359296, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv4/conv4.12/conv/conv.0/Conv"](%/conv4/conv4.11/conv/conv.6/Conv_output_0, %onnx::Conv_578, %onnx::Conv_579), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.12/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.12/conv/conv.2/Relu_output_0 : Float(1, 576, 64, 64, strides=[2359296, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv4/conv4.12/conv/conv.2/Relu"](%/conv4/conv4.12/conv/conv.0/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.12/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.2 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv4/conv4.12/conv/conv.3/Conv_output_0 : Float(1, 576, 64, 64, strides=[2359296, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=576, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/conv4/conv4.12/conv/conv.3/Conv"](%/conv4/conv4.12/conv/conv.2/Relu_output_0, %onnx::Conv_581, %onnx::Conv_582), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.12/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.12/conv/conv.5/Relu_output_0 : Float(1, 576, 64, 64, strides=[2359296, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv4/conv4.12/conv/conv.5/Relu"](%/conv4/conv4.12/conv/conv.3/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.12/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.5 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv4/conv4.12/conv/conv.6/Conv_output_0 : Float(1, 96, 64, 64, strides=[393216, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv4/conv4.12/conv/conv.6/Conv"](%/conv4/conv4.12/conv/conv.5/Relu_output_0, %onnx::Conv_584, %onnx::Conv_585), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.12/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.12/Add_output_0 : Float(1, 96, 64, 64, strides=[393216, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Add[onnx_name="/conv4/conv4.12/Add"](%/conv4/conv4.11/conv/conv.6/Conv_output_0, %/conv4/conv4.12/conv/conv.6/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.12 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/ao/nn/quantized/modules/functional_modules.py:45:0
  %/conv4/conv4.13/conv/conv.0/Conv_output_0 : Float(1, 576, 64, 64, strides=[2359296, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv4/conv4.13/conv/conv.0/Conv"](%/conv4/conv4.12/Add_output_0, %onnx::Conv_587, %onnx::Conv_588), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.13/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.13/conv/conv.2/Relu_output_0 : Float(1, 576, 64, 64, strides=[2359296, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv4/conv4.13/conv/conv.2/Relu"](%/conv4/conv4.13/conv/conv.0/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.13/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.2 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv4/conv4.13/conv/conv.3/Conv_output_0 : Float(1, 576, 64, 64, strides=[2359296, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=576, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/conv4/conv4.13/conv/conv.3/Conv"](%/conv4/conv4.13/conv/conv.2/Relu_output_0, %onnx::Conv_590, %onnx::Conv_591), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.13/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.13/conv/conv.5/Relu_output_0 : Float(1, 576, 64, 64, strides=[2359296, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/conv4/conv4.13/conv/conv.5/Relu"](%/conv4/conv4.13/conv/conv.3/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.13/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.5 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/conv4/conv4.13/conv/conv.6/Conv_output_0 : Float(1, 96, 64, 64, strides=[393216, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv4/conv4.13/conv/conv.6/Conv"](%/conv4/conv4.13/conv/conv.5/Relu_output_0, %onnx::Conv_593, %onnx::Conv_594), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.13/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/conv4/conv4.13/Add_output_0 : Float(1, 96, 64, 64, strides=[393216, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Add[onnx_name="/conv4/conv4.13/Add"](%/conv4/conv4.12/Add_output_0, %/conv4/conv4.13/conv/conv.6/Conv_output_0), scope: m2unet.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet.InvertedResidual::conv4.13 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/ao/nn/quantized/modules/functional_modules.py:45:0
  %/decode4/upsample/Constant_output_0 : Float(4, strides=[1], requires_grad=0, device=cpu) = onnx::Constant[value= 1  1  2  2 [ CPUFloatType{4} ], onnx_name="/decode4/upsample/Constant"](), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode4/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:3919:0
  %onnx::Resize_424 : Tensor? = prim::Constant(), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode4/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:3919:0
  %/decode4/upsample/Resize_output_0 : Float(1, 96, 128, 128, strides=[1572864, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Resize[coordinate_transformation_mode="half_pixel", cubic_coeff_a=-0.75, mode="linear", nearest_mode="floor", onnx_name="/decode4/upsample/Resize"](%/conv4/conv4.13/Add_output_0, %onnx::Resize_424, %/decode4/upsample/Constant_output_0), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode4/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:3919:0
  %/decode4/Concat_output_0 : Float(1, 128, 128, 128, strides=[2097152, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Concat[axis=1, onnx_name="/decode4/Concat"](%/decode4/upsample/Resize_output_0, %/conv3/conv3.6/Add_output_0), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode4 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/ao/nn/quantized/modules/functional_modules.py:71:0
  %/decode4/ir1/conv/conv.0/Conv_output_0 : Float(1, 19, 128, 128, strides=[311296, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/decode4/ir1/conv/conv.0/Conv"](%/decode4/Concat_output_0, %onnx::Conv_596, %onnx::Conv_597), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode4/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/decode4/ir1/conv/conv.2/Relu_output_0 : Float(1, 19, 128, 128, strides=[311296, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/decode4/ir1/conv/conv.2/Relu"](%/decode4/ir1/conv/conv.0/Conv_output_0), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode4/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.2 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/decode4/ir1/conv/conv.3/Conv_output_0 : Float(1, 19, 128, 128, strides=[311296, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=19, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/decode4/ir1/conv/conv.3/Conv"](%/decode4/ir1/conv/conv.2/Relu_output_0, %onnx::Conv_599, %onnx::Conv_600), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode4/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/decode4/ir1/conv/conv.5/Relu_output_0 : Float(1, 19, 128, 128, strides=[311296, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/decode4/ir1/conv/conv.5/Relu"](%/decode4/ir1/conv/conv.3/Conv_output_0), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode4/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.5 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/decode4/ir1/conv/conv.6/Conv_output_0 : Float(1, 64, 128, 128, strides=[1048576, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/decode4/ir1/conv/conv.6/Conv"](%/decode4/ir1/conv/conv.5/Relu_output_0, %onnx::Conv_602, %onnx::Conv_603), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode4/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/decode3/upsample/Constant_output_0 : Float(4, strides=[1], requires_grad=0, device=cpu) = onnx::Constant[value= 1  1  2  2 [ CPUFloatType{4} ], onnx_name="/decode3/upsample/Constant"](), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode3/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:3919:0
  %onnx::Resize_438 : Tensor? = prim::Constant(), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode3/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:3919:0
  %/decode3/upsample/Resize_output_0 : Float(1, 64, 256, 256, strides=[4194304, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Resize[coordinate_transformation_mode="half_pixel", cubic_coeff_a=-0.75, mode="linear", nearest_mode="floor", onnx_name="/decode3/upsample/Resize"](%/decode4/ir1/conv/conv.6/Conv_output_0, %onnx::Resize_438, %/decode3/upsample/Constant_output_0), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode3/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:3919:0
  %/decode3/Concat_output_0 : Float(1, 88, 256, 256, strides=[5767168, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Concat[axis=1, onnx_name="/decode3/Concat"](%/decode3/upsample/Resize_output_0, %/conv2/conv2.3/Add_output_0), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode3 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/ao/nn/quantized/modules/functional_modules.py:71:0
  %/decode3/ir1/conv/conv.0/Conv_output_0 : Float(1, 13, 256, 256, strides=[851968, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/decode3/ir1/conv/conv.0/Conv"](%/decode3/Concat_output_0, %onnx::Conv_605, %onnx::Conv_606), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode3/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/decode3/ir1/conv/conv.2/Relu_output_0 : Float(1, 13, 256, 256, strides=[851968, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/decode3/ir1/conv/conv.2/Relu"](%/decode3/ir1/conv/conv.0/Conv_output_0), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode3/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.2 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/decode3/ir1/conv/conv.3/Conv_output_0 : Float(1, 13, 256, 256, strides=[851968, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=13, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/decode3/ir1/conv/conv.3/Conv"](%/decode3/ir1/conv/conv.2/Relu_output_0, %onnx::Conv_608, %onnx::Conv_609), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode3/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/decode3/ir1/conv/conv.5/Relu_output_0 : Float(1, 13, 256, 256, strides=[851968, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/decode3/ir1/conv/conv.5/Relu"](%/decode3/ir1/conv/conv.3/Conv_output_0), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode3/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.5 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/decode3/ir1/conv/conv.6/Conv_output_0 : Float(1, 44, 256, 256, strides=[2883584, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/decode3/ir1/conv/conv.6/Conv"](%/decode3/ir1/conv/conv.5/Relu_output_0, %onnx::Conv_611, %onnx::Conv_612), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode3/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/decode2/upsample/Constant_output_0 : Float(4, strides=[1], requires_grad=0, device=cpu) = onnx::Constant[value= 1  1  2  2 [ CPUFloatType{4} ], onnx_name="/decode2/upsample/Constant"](), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode2/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:3919:0
  %onnx::Resize_452 : Tensor? = prim::Constant(), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode2/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:3919:0
  %/decode2/upsample/Resize_output_0 : Float(1, 44, 512, 512, strides=[11534336, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Resize[coordinate_transformation_mode="half_pixel", cubic_coeff_a=-0.75, mode="linear", nearest_mode="floor", onnx_name="/decode2/upsample/Resize"](%/decode3/ir1/conv/conv.6/Conv_output_0, %onnx::Resize_452, %/decode2/upsample/Constant_output_0), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode2/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:3919:0
  %/decode2/Concat_output_0 : Float(1, 60, 512, 512, strides=[15728640, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Concat[axis=1, onnx_name="/decode2/Concat"](%/decode2/upsample/Resize_output_0, %/conv1/conv1.1/conv/conv.3/Conv_output_0), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode2 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/ao/nn/quantized/modules/functional_modules.py:71:0
  %/decode2/ir1/conv/conv.0/Conv_output_0 : Float(1, 9, 512, 512, strides=[2359296, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/decode2/ir1/conv/conv.0/Conv"](%/decode2/Concat_output_0, %onnx::Conv_614, %onnx::Conv_615), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode2/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/decode2/ir1/conv/conv.2/Relu_output_0 : Float(1, 9, 512, 512, strides=[2359296, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/decode2/ir1/conv/conv.2/Relu"](%/decode2/ir1/conv/conv.0/Conv_output_0), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode2/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.2 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/decode2/ir1/conv/conv.3/Conv_output_0 : Float(1, 9, 512, 512, strides=[2359296, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=9, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/decode2/ir1/conv/conv.3/Conv"](%/decode2/ir1/conv/conv.2/Relu_output_0, %onnx::Conv_617, %onnx::Conv_618), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode2/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/decode2/ir1/conv/conv.5/Relu_output_0 : Float(1, 9, 512, 512, strides=[2359296, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/decode2/ir1/conv/conv.5/Relu"](%/decode2/ir1/conv/conv.3/Conv_output_0), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode2/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.5 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/decode2/ir1/conv/conv.6/Conv_output_0 : Float(1, 30, 512, 512, strides=[7864320, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/decode2/ir1/conv/conv.6/Conv"](%/decode2/ir1/conv/conv.5/Relu_output_0, %onnx::Conv_620, %onnx::Conv_621), scope: m2unet.M2UNet::/m2unet.DecoderBlock::decode2/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/decode1/upsample/Constant_output_0 : Float(4, strides=[1], requires_grad=0, device=cpu) = onnx::Constant[value= 1  1  2  2 [ CPUFloatType{4} ], onnx_name="/decode1/upsample/Constant"](), scope: m2unet.M2UNet::/m2unet.LastDecoderBlock::decode1/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:3919:0
  %onnx::Resize_466 : Tensor? = prim::Constant(), scope: m2unet.M2UNet::/m2unet.LastDecoderBlock::decode1/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:3919:0
  %/decode1/upsample/Resize_output_0 : Float(1, 30, 1024, 1024, strides=[31457280, 1048576, 1024, 1], requires_grad=1, device=cuda:0) = onnx::Resize[coordinate_transformation_mode="half_pixel", cubic_coeff_a=-0.75, mode="linear", nearest_mode="floor", onnx_name="/decode1/upsample/Resize"](%/decode2/ir1/conv/conv.6/Conv_output_0, %onnx::Resize_466, %/decode1/upsample/Constant_output_0), scope: m2unet.M2UNet::/m2unet.LastDecoderBlock::decode1/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:3919:0
  %/decode1/Concat_output_0 : Float(1, 31, 1024, 1024, strides=[32505856, 1048576, 1024, 1], requires_grad=1, device=cuda:0) = onnx::Concat[axis=1, onnx_name="/decode1/Concat"](%/decode1/upsample/Resize_output_0, %input.1), scope: m2unet.M2UNet::/m2unet.LastDecoderBlock::decode1 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/ao/nn/quantized/modules/functional_modules.py:71:0
  %/decode1/ir1/conv/conv.0/Conv_output_0 : Float(1, 5, 1024, 1024, strides=[5242880, 1048576, 1024, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/decode1/ir1/conv/conv.0/Conv"](%/decode1/Concat_output_0, %onnx::Conv_623, %onnx::Conv_624), scope: m2unet.M2UNet::/m2unet.LastDecoderBlock::decode1/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/decode1/ir1/conv/conv.2/Relu_output_0 : Float(1, 5, 1024, 1024, strides=[5242880, 1048576, 1024, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/decode1/ir1/conv/conv.2/Relu"](%/decode1/ir1/conv/conv.0/Conv_output_0), scope: m2unet.M2UNet::/m2unet.LastDecoderBlock::decode1/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.2 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/decode1/ir1/conv/conv.3/Conv_output_0 : Float(1, 5, 1024, 1024, strides=[5242880, 1048576, 1024, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=5, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/decode1/ir1/conv/conv.3/Conv"](%/decode1/ir1/conv/conv.2/Relu_output_0, %onnx::Conv_626, %onnx::Conv_627), scope: m2unet.M2UNet::/m2unet.LastDecoderBlock::decode1/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %/decode1/ir1/conv/conv.5/Relu_output_0 : Float(1, 5, 1024, 1024, strides=[5242880, 1048576, 1024, 1], requires_grad=1, device=cuda:0) = onnx::Relu[onnx_name="/decode1/ir1/conv/conv.5/Relu"](%/decode1/ir1/conv/conv.3/Conv_output_0), scope: m2unet.M2UNet::/m2unet.LastDecoderBlock::decode1/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU::conv.5 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/functional.py:1453:0
  %/decode1/ir1/conv/conv.6/Conv_output_0 : Float(1, 16, 1024, 1024, strides=[16777216, 1048576, 1024, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/decode1/ir1/conv/conv.6/Conv"](%/decode1/ir1/conv/conv.5/Relu_output_0, %onnx::Conv_629, %onnx::Conv_630), scope: m2unet.M2UNet::/m2unet.LastDecoderBlock::decode1/m2unet.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  %477 : Float(1, 1, 1024, 1024, strides=[1048576, 1048576, 1024, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/decode1/conv/conv.0/Conv"](%/decode1/ir1/conv/conv.6/Conv_output_0, %decode1.conv.0.weight, %decode1.conv.0.bias), scope: m2unet.M2UNet::/m2unet.LastDecoderBlock::decode1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/miniconda3/envs/tensorRT/lib/python3.10/site-packages/torch/nn/modules/conv.py:458:0
  return (%477)

However, it did produce an onnx file as an output.

I then ran trtexec --onnx=models/flat_model_56_10.onnx --saveEngine=models/flat_model_56_10_from_onnx.trt --explicitBatch to convert from onnx to trt on my Jetson Xavier NX. I got the following logs:

&&&& RUNNING TensorRT.trtexec [TensorRT v8401] # trtexec --onnx=models/flat_model_56_10.onnx --saveEngine=models/flat_model_56_10_from_onnx.trt --explicitBatch
[03/16/2023-15:48:13] [W] --explicitBatch flag has been deprecated and has no effect!
[03/16/2023-15:48:13] [W] Explicit batch dim is automatically enabled if input model is ONNX or if dynamic shapes are provided when the engine is built.
[03/16/2023-15:48:13] [I] === Model Options ===
[03/16/2023-15:48:13] [I] Format: ONNX
[03/16/2023-15:48:13] [I] Model: /media/prakashlab/SSD/quantize_m2unet/models/flat_model_56_10.onnx
[03/16/2023-15:48:13] [I] Output:
[03/16/2023-15:48:13] [I] === Build Options ===
[03/16/2023-15:48:13] [I] Max batch: explicit batch
[03/16/2023-15:48:13] [I] Memory Pools: workspace: default, dlaSRAM: default, dlaLocalDRAM: default, dlaGlobalDRAM: default
[03/16/2023-15:48:13] [I] minTiming: 1
[03/16/2023-15:48:13] [I] avgTiming: 8
[03/16/2023-15:48:13] [I] Precision: FP32
[03/16/2023-15:48:13] [I] LayerPrecisions: 
[03/16/2023-15:48:13] [I] Calibration: 
[03/16/2023-15:48:13] [I] Refit: Disabled
[03/16/2023-15:48:13] [I] Sparsity: Disabled
[03/16/2023-15:48:13] [I] Safe mode: Disabled
[03/16/2023-15:48:13] [I] DirectIO mode: Disabled
[03/16/2023-15:48:13] [I] Restricted mode: Disabled
[03/16/2023-15:48:13] [I] Build only: Disabled
[03/16/2023-15:48:13] [I] Save engine: /media/prakashlab/SSD/quantize_m2unet/models/flat_model_56_10_from_onnx.trt
[03/16/2023-15:48:13] [I] Load engine: 
[03/16/2023-15:48:13] [I] Profiling verbosity: 0
[03/16/2023-15:48:13] [I] Tactic sources: Using default tactic sources
[03/16/2023-15:48:13] [I] timingCacheMode: local
[03/16/2023-15:48:13] [I] timingCacheFile: 
[03/16/2023-15:48:13] [I] Input(s)s format: fp32:CHW
[03/16/2023-15:48:13] [I] Output(s)s format: fp32:CHW
[03/16/2023-15:48:13] [I] Input build shapes: model
[03/16/2023-15:48:13] [I] Input calibration shapes: model
[03/16/2023-15:48:13] [I] === System Options ===
[03/16/2023-15:48:13] [I] Device: 0
[03/16/2023-15:48:13] [I] DLACore: 
[03/16/2023-15:48:13] [I] Plugins:
[03/16/2023-15:48:13] [I] === Inference Options ===
[03/16/2023-15:48:13] [I] Batch: Explicit
[03/16/2023-15:48:13] [I] Input inference shapes: model
[03/16/2023-15:48:13] [I] Iterations: 10
[03/16/2023-15:48:13] [I] Duration: 3s (+ 200ms warm up)
[03/16/2023-15:48:13] [I] Sleep time: 0ms
[03/16/2023-15:48:13] [I] Idle time: 0ms
[03/16/2023-15:48:13] [I] Streams: 1
[03/16/2023-15:48:13] [I] ExposeDMA: Disabled
[03/16/2023-15:48:13] [I] Data transfers: Enabled
[03/16/2023-15:48:13] [I] Spin-wait: Disabled
[03/16/2023-15:48:13] [I] Multithreading: Disabled
[03/16/2023-15:48:13] [I] CUDA Graph: Disabled
[03/16/2023-15:48:13] [I] Separate profiling: Disabled
[03/16/2023-15:48:13] [I] Time Deserialize: Disabled
[03/16/2023-15:48:13] [I] Time Refit: Disabled
[03/16/2023-15:48:13] [I] Inputs:
[03/16/2023-15:48:13] [I] === Reporting Options ===
[03/16/2023-15:48:13] [I] Verbose: Disabled
[03/16/2023-15:48:13] [I] Averages: 10 inferences
[03/16/2023-15:48:13] [I] Percentile: 99
[03/16/2023-15:48:13] [I] Dump refittable layers:Disabled
[03/16/2023-15:48:13] [I] Dump output: Disabled
[03/16/2023-15:48:13] [I] Profile: Disabled
[03/16/2023-15:48:13] [I] Export timing to JSON file: 
[03/16/2023-15:48:13] [I] Export output to JSON file: 
[03/16/2023-15:48:13] [I] Export profile to JSON file: 
[03/16/2023-15:48:13] [I] 
[03/16/2023-15:48:14] [I] === Device Information ===
[03/16/2023-15:48:14] [I] Selected Device: Xavier
[03/16/2023-15:48:14] [I] Compute Capability: 7.2
[03/16/2023-15:48:14] [I] SMs: 6
[03/16/2023-15:48:14] [I] Compute Clock Rate: 1.109 GHz
[03/16/2023-15:48:14] [I] Device Global Memory: 6846 MiB
[03/16/2023-15:48:14] [I] Shared Memory per SM: 96 KiB
[03/16/2023-15:48:14] [I] Memory Bus Width: 256 bits (ECC disabled)
[03/16/2023-15:48:14] [I] Memory Clock Rate: 1.109 GHz
[03/16/2023-15:48:14] [I] 
[03/16/2023-15:48:14] [I] TensorRT version: 8.4.1
[03/16/2023-15:48:15] [I] [TRT] [MemUsageChange] Init CUDA: CPU +185, GPU +0, now: CPU 209, GPU 3118 (MiB)
[03/16/2023-15:48:19] [I] [TRT] [MemUsageChange] Init builder kernel library: CPU +131, GPU +124, now: CPU 359, GPU 3255 (MiB)
[03/16/2023-15:48:19] [I] Start parsing network model
[03/16/2023-15:48:19] [I] [TRT] ----------------------------------------------------------------
[03/16/2023-15:48:19] [I] [TRT] Input filename:   /media/prakashlab/SSD/quantize_m2unet/models/flat_model_56_10.onnx
[03/16/2023-15:48:19] [I] [TRT] ONNX IR version:  0.0.7
[03/16/2023-15:48:19] [I] [TRT] Opset version:    12
[03/16/2023-15:48:19] [I] [TRT] Producer name:    pytorch
[03/16/2023-15:48:19] [I] [TRT] Producer version: 1.13.0
[03/16/2023-15:48:19] [I] [TRT] Domain:           
[03/16/2023-15:48:19] [I] [TRT] Model version:    0
[03/16/2023-15:48:19] [I] [TRT] Doc string:       
[03/16/2023-15:48:19] [I] [TRT] ----------------------------------------------------------------
[03/16/2023-15:48:19] [I] Finish parsing network model
[03/16/2023-15:48:20] [I] [TRT] ---------- Layers Running on DLA ----------
[03/16/2023-15:48:20] [I] [TRT] ---------- Layers Running on GPU ----------
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] COPY: Reformatting CopyNode for Network Input input.1
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv1/conv1.0/conv1.0.0/Conv + /conv1/conv1.0/conv1.0.2/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv1/conv1.1/conv/conv.0/Conv + /conv1/conv1.1/conv/conv.2/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv1/conv1.1/conv/conv.3/Conv
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv2/conv2.2/conv/conv.0/Conv + /conv2/conv2.2/conv/conv.2/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv2/conv2.2/conv/conv.3/Conv + /conv2/conv2.2/conv/conv.5/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv2/conv2.2/conv/conv.6/Conv
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv2/conv2.3/conv/conv.0/Conv + /conv2/conv2.3/conv/conv.2/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv2/conv2.3/conv/conv.3/Conv + /conv2/conv2.3/conv/conv.5/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv2/conv2.3/conv/conv.6/Conv + /conv2/conv2.3/Add
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv3/conv3.4/conv/conv.0/Conv + /conv3/conv3.4/conv/conv.2/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv3/conv3.4/conv/conv.3/Conv + /conv3/conv3.4/conv/conv.5/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv3/conv3.4/conv/conv.6/Conv
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv3/conv3.5/conv/conv.0/Conv + /conv3/conv3.5/conv/conv.2/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv3/conv3.5/conv/conv.3/Conv + /conv3/conv3.5/conv/conv.5/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv3/conv3.5/conv/conv.6/Conv + /conv3/conv3.5/Add
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv3/conv3.6/conv/conv.0/Conv + /conv3/conv3.6/conv/conv.2/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv3/conv3.6/conv/conv.3/Conv + /conv3/conv3.6/conv/conv.5/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv3/conv3.6/conv/conv.6/Conv + /conv3/conv3.6/Add
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.7/conv/conv.0/Conv + /conv4/conv4.7/conv/conv.2/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.7/conv/conv.3/Conv + /conv4/conv4.7/conv/conv.5/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.7/conv/conv.6/Conv
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.8/conv/conv.0/Conv + /conv4/conv4.8/conv/conv.2/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.8/conv/conv.3/Conv + /conv4/conv4.8/conv/conv.5/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.8/conv/conv.6/Conv + /conv4/conv4.8/Add
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.9/conv/conv.0/Conv + /conv4/conv4.9/conv/conv.2/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.9/conv/conv.3/Conv + /conv4/conv4.9/conv/conv.5/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.9/conv/conv.6/Conv + /conv4/conv4.9/Add
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.10/conv/conv.0/Conv + /conv4/conv4.10/conv/conv.2/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.10/conv/conv.3/Conv + /conv4/conv4.10/conv/conv.5/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.10/conv/conv.6/Conv + /conv4/conv4.10/Add
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.11/conv/conv.0/Conv + /conv4/conv4.11/conv/conv.2/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.11/conv/conv.3/Conv + /conv4/conv4.11/conv/conv.5/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.11/conv/conv.6/Conv
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.12/conv/conv.0/Conv + /conv4/conv4.12/conv/conv.2/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.12/conv/conv.3/Conv + /conv4/conv4.12/conv/conv.5/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.12/conv/conv.6/Conv + /conv4/conv4.12/Add
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.13/conv/conv.0/Conv + /conv4/conv4.13/conv/conv.2/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.13/conv/conv.3/Conv + /conv4/conv4.13/conv/conv.5/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /conv4/conv4.13/conv/conv.6/Conv + /conv4/conv4.13/Add
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] RESIZE: /decode4/upsample/Resize
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] COPY: /decode4/upsample/Resize_output_0 copy
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] COPY: /conv3/conv3.6/Add_output_0 copy
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /decode4/ir1/conv/conv.0/Conv + /decode4/ir1/conv/conv.2/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /decode4/ir1/conv/conv.3/Conv + /decode4/ir1/conv/conv.5/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /decode4/ir1/conv/conv.6/Conv
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] RESIZE: /decode3/upsample/Resize
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] COPY: /decode3/upsample/Resize_output_0 copy
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] COPY: /conv2/conv2.3/Add_output_0 copy
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /decode3/ir1/conv/conv.0/Conv + /decode3/ir1/conv/conv.2/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /decode3/ir1/conv/conv.3/Conv + /decode3/ir1/conv/conv.5/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /decode3/ir1/conv/conv.6/Conv
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] RESIZE: /decode2/upsample/Resize
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] COPY: /decode2/upsample/Resize_output_0 copy
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /decode2/ir1/conv/conv.0/Conv + /decode2/ir1/conv/conv.2/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /decode2/ir1/conv/conv.3/Conv + /decode2/ir1/conv/conv.5/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /decode2/ir1/conv/conv.6/Conv
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] RESIZE: /decode1/upsample/Resize
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] COPY: /decode1/upsample/Resize_output_0 copy
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] COPY: input.1 copy
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /decode1/ir1/conv/conv.0/Conv + /decode1/ir1/conv/conv.2/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /decode1/ir1/conv/conv.3/Conv + /decode1/ir1/conv/conv.5/Relu
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /decode1/ir1/conv/conv.6/Conv
[03/16/2023-15:48:20] [I] [TRT] [GpuLayer] CONVOLUTION: /decode1/conv/conv.0/Conv
[03/16/2023-15:48:22] [I] [TRT] [MemUsageChange] Init cuBLAS/cuBLASLt: CPU +260, GPU +241, now: CPU 622, GPU 3503 (MiB)
[03/16/2023-15:48:23] [I] [TRT] [MemUsageChange] Init cuDNN: CPU +86, GPU +82, now: CPU 708, GPU 3585 (MiB)
[03/16/2023-15:48:23] [I] [TRT] Local timing cache in use. Profiling results in this builder pass will not be stored.
[03/16/2023-15:53:35] [I] [TRT] Detected 1 inputs and 1 output network tensors.
[03/16/2023-15:53:35] [I] [TRT] Total Host Persistent Memory: 84288
[03/16/2023-15:53:35] [I] [TRT] Total Device Persistent Memory: 23929856
[03/16/2023-15:53:35] [I] [TRT] Total Scratch Memory: 4194304
[03/16/2023-15:53:35] [I] [TRT] [MemUsageStats] Peak memory usage of TRT CPU/GPU memory allocators: CPU 0 MiB, GPU 1866 MiB
[03/16/2023-15:53:35] [I] [TRT] [BlockAssignment] Algorithm ShiftNTopDown took 17.6213ms to assign 8 blocks to 60 nodes requiring 309854212 bytes.
[03/16/2023-15:53:35] [I] [TRT] Total Activation Memory: 309854212
[03/16/2023-15:53:35] [I] [TRT] [MemUsageChange] Init cuDNN: CPU +1, GPU +0, now: CPU 973, GPU 4423 (MiB)
[03/16/2023-15:53:35] [I] [TRT] [MemUsageChange] TensorRT-managed allocation in building engine: CPU +0, GPU +26, now: CPU 0, GPU 26 (MiB)
[03/16/2023-15:53:35] [W] [TRT] The getMaxBatchSize() function should not be used with an engine built from a network created with NetworkDefinitionCreationFlag::kEXPLICIT_BATCH flag. This function will always return 1.
[03/16/2023-15:53:35] [W] [TRT] The getMaxBatchSize() function should not be used with an engine built from a network created with NetworkDefinitionCreationFlag::kEXPLICIT_BATCH flag. This function will always return 1.
[03/16/2023-15:53:35] [I] Engine built in 321.884 sec.
[03/16/2023-15:53:35] [I] [TRT] [MemUsageChange] Init CUDA: CPU +0, GPU +0, now: CPU 841, GPU 4379 (MiB)
[03/16/2023-15:53:35] [I] [TRT] Loaded engine size: 2 MiB
[03/16/2023-15:53:35] [I] [TRT] [MemUsageChange] Init cuDNN: CPU +0, GPU +0, now: CPU 841, GPU 4379 (MiB)
[03/16/2023-15:53:35] [I] [TRT] [MemUsageChange] TensorRT-managed allocation in engine deserialization: CPU +0, GPU +24, now: CPU 0, GPU 24 (MiB)
[03/16/2023-15:53:35] [I] Engine deserialized in 0.0209206 sec.
[03/16/2023-15:53:35] [I] [TRT] [MemUsageChange] Init cuDNN: CPU +0, GPU +0, now: CPU 841, GPU 4379 (MiB)
[03/16/2023-15:53:35] [I] [TRT] [MemUsageChange] TensorRT-managed allocation in IExecutionContext creation: CPU +0, GPU +319, now: CPU 0, GPU 343 (MiB)
[03/16/2023-15:53:35] [I] Using random values for input input.1
[03/16/2023-15:53:36] [I] Created input binding for input.1 with dimensions 1x1x1024x1024
[03/16/2023-15:53:36] [I] Using random values for output 477
[03/16/2023-15:53:36] [I] Created output binding for 477 with dimensions 1x1x1024x1024
[03/16/2023-15:53:36] [I] Starting inference
[03/16/2023-15:53:39] [I] Warmup completed 4 queries over 200 ms
[03/16/2023-15:53:39] [I] Timing trace has 51 queries over 3.16491 s
[03/16/2023-15:53:39] [I] 
[03/16/2023-15:53:39] [I] === Trace details ===
[03/16/2023-15:53:39] [I] Trace averages of 10 runs:
[03/16/2023-15:53:39] [I] Average on 10 runs - GPU latency: 60.9328 ms - Host latency: 61.633 ms (enqueue 1.44854 ms)
[03/16/2023-15:53:39] [I] Average on 10 runs - GPU latency: 60.8995 ms - Host latency: 61.5924 ms (enqueue 1.40194 ms)
[03/16/2023-15:53:39] [I] Average on 10 runs - GPU latency: 60.8387 ms - Host latency: 61.5485 ms (enqueue 1.43955 ms)
[03/16/2023-15:53:39] [I] Average on 10 runs - GPU latency: 60.8859 ms - Host latency: 61.5922 ms (enqueue 1.3922 ms)
[03/16/2023-15:53:39] [I] Average on 10 runs - GPU latency: 60.8665 ms - Host latency: 61.5735 ms (enqueue 1.39307 ms)
[03/16/2023-15:53:39] [I] 
[03/16/2023-15:53:39] [I] === Performance summary ===
[03/16/2023-15:53:39] [I] Throughput: 16.1142 qps
[03/16/2023-15:53:39] [I] Latency: min = 60.8 ms, max = 63.9493 ms, mean = 61.5725 ms, median = 61.3909 ms, percentile(99%) = 63.9493 ms
[03/16/2023-15:53:39] [I] Enqueue Time: min = 1.23022 ms, max = 1.6532 ms, mean = 1.41308 ms, median = 1.39172 ms, percentile(99%) = 1.6532 ms
[03/16/2023-15:53:39] [I] H2D Latency: min = 0.286621 ms, max = 0.343994 ms, mean = 0.299214 ms, median = 0.295532 ms, percentile(99%) = 0.343994 ms
[03/16/2023-15:53:39] [I] GPU Compute Time: min = 60.3404 ms, max = 63.2629 ms, mean = 60.8743 ms, median = 60.6898 ms, percentile(99%) = 63.2629 ms
[03/16/2023-15:53:39] [I] D2H Latency: min = 0.156006 ms, max = 0.428345 ms, mean = 0.398971 ms, median = 0.402588 ms, percentile(99%) = 0.428345 ms
[03/16/2023-15:53:39] [I] Total Host Walltime: 3.16491 s
[03/16/2023-15:53:39] [I] Total GPU Compute Time: 3.10459 s
[03/16/2023-15:53:39] [W] * GPU compute time is unstable, with coefficient of variance = 1.23481%.
[03/16/2023-15:53:39] [W]   If not already in use, locking GPU clock frequency or adding --useSpinWait may improve the stability.
[03/16/2023-15:53:39] [I] Explanations of the performance metrics are printed in the verbose logs.
[03/16/2023-15:53:39] [I] 
&&&& PASSED TensorRT.trtexec [TensorRT v8401] # trtexec --onnx=models/flat_model_56_10.onnx --saveEngine=models/flat_model_56_10_from_onnx.trt --explicitBatch

I tried running the model in Python using the following code:

import torch
import torch2trt
import numpy as np

def load_trt_model(path):
    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
    model_trt = torch2trt.TRTModule()
    model_trt.load_state_dict(torch.load(path))

    return device, model_trt

def process_image_for_trt(image, device):
    '''
    Takes a monochrome image (x, y) 
    and returns an object with the correct data format
    for running inference. 
    '''
    x = np.expand_dims(image, 0)
    x = np.expand_dims(x, 0)
    x = torch.from_numpy(x).float()
    x = x.to(device)
    return x

def inference_on_image(model, image, device):
    '''
    Take an image and model and return the inference result
    as a np array
    '''
    im = process_image_for_trt(image, device)
    out = model(im)[0].detach().cpu().float().numpy()
    return out

This code produced nonsensical outputs; instead of performing segmentation, it distorts the input in uninformative ways. Overall it looks like I am not seeing the same error as in torch2trt; I am not sure why this method is failing.

AastaLLL · March 23, 2023, 3:31am

Hi,

The shape inference of prim::Constant type is missing, so it may result in wrong shape inference for the exported graph.

Based on the log, there are some issues when the ONNX file is generated.
Which opset you used when creating the file?

We found some similar errors online and they can fix the issue with opset=11.
Could you also give it a try?

Thanks.

kmarx · March 23, 2023, 3:34am

Thank you for the advice, I had been using opset 12. I will give opset 11 a try.

Do you have suggestions for using torch2trt? If possible, I’d like to convert directly from torch to TensorRT rather than using onnx.

kmarx · March 23, 2023, 5:38pm

It would appear opset=11 did not help.

Here’s the log when creating the .onnx model. I had to exclude some of the logs to fit in the post word limit:

Exported graph: graph(%input.1 : Float(1, 1, 1024, 1024, strides=[1048576, 1048576, 1024, 1], requires_grad=0, device=cuda:0),
      %decode1.conv.0.weight : Float(1, 16, 1, 1, strides=[16, 1, 1, 1], requires_grad=1, device=cuda:0),
      %decode1.conv.0.bias : Float(1, strides=[1], requires_grad=1, device=cuda:0),
      %onnx::Conv_547 : Float(32, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_548 : Float(32, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_550 : Float(32, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_551 : Float(32, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_553 : Float(16, 32, 1, 1, strides=[32, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_554 : Float(16, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_556 : Float(96, 16, 1, 1, strides=[16, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_557 : Float(96, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_559 : Float(96, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_560 : Float(96, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_562 : Float(24, 96, 1, 1, strides=[96, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_563 : Float(24, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_565 : Float(144, 24, 1, 1, strides=[24, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_566 : Float(144, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_568 : Float(144, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_569 : Float(144, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_571 : Float(24, 144, 1, 1, strides=[144, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_572 : Float(24, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_574 : Float(144, 24, 1, 1, strides=[24, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_575 : Float(144, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_577 : Float(144, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_578 : Float(144, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_580 : Float(32, 144, 1, 1, strides=[144, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_581 : Float(32, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_583 : Float(192, 32, 1, 1, strides=[32, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_584 : Float(192, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_586 : Float(192, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_587 : Float(192, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_589 : Float(32, 192, 1, 1, strides=[192, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_590 : Float(32, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_592 : Float(192, 32, 1, 1, strides=[32, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_593 : Float(192, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_595 : Float(192, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_596 : Float(192, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_598 : Float(32, 192, 1, 1, strides=[192, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_599 : Float(32, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_601 : Float(192, 32, 1, 1, strides=[32, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_602 : Float(192, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_604 : Float(192, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_605 : Float(192, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_607 : Float(64, 192, 1, 1, strides=[192, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_608 : Float(64, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_610 : Float(384, 64, 1, 1, strides=[64, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_611 : Float(384, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_613 : Float(384, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_614 : Float(384, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_616 : Float(64, 384, 1, 1, strides=[384, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_617 : Float(64, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_620 : Float(384, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_623 : Float(384, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_629 : Float(384, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_632 : Float(384, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_635 : Float(64, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_638 : Float(384, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_641 : Float(384, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_643 : Float(96, 384, 1, 1, strides=[384, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_644 : Float(96, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_646 : Float(576, 96, 1, 1, strides=[96, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_647 : Float(576, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_649 : Float(576, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_650 : Float(576, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_652 : Float(96, 576, 1, 1, strides=[576, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_656 : Float(576, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_659 : Float(576, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_664 : Float(19, 128, 1, 1, strides=[128, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_665 : Float(19, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_667 : Float(19, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_668 : Float(19, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_670 : Float(64, 19, 1, 1, strides=[19, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_671 : Float(64, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_673 : Float(13, 88, 1, 1, strides=[88, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_674 : Float(13, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_676 : Float(13, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_677 : Float(13, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_679 : Float(44, 13, 1, 1, strides=[13, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_680 : Float(44, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_682 : Float(9, 60, 1, 1, strides=[60, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_683 : Float(9, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_685 : Float(9, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_686 : Float(9, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_688 : Float(30, 9, 1, 1, strides=[9, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_689 : Float(30, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_691 : Float(5, 31, 1, 1, strides=[31, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_692 : Float(5, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_694 : Float(5, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_695 : Float(5, strides=[1], requires_grad=0, device=cuda:0),
      %onnx::Conv_697 : Float(16, 5, 1, 1, strides=[5, 1, 1, 1], requires_grad=0, device=cuda:0),
      %onnx::Conv_698 : Float(16, strides=[1], requires_grad=0, device=cuda:0)):
  %onnx::Conv_662 : Float(96, strides=[1], requires_grad=0, device=cuda:0) = onnx::Identity(%onnx::Conv_644)
  %onnx::Conv_661 : Float(96, 576, 1, 1, strides=[576, 1, 1, 1], requires_grad=0, device=cuda:0) = onnx::Identity(%onnx::Conv_652)
  %onnx::Conv_658 : Float(576, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0) = onnx::Identity(%onnx::Conv_649)
  %onnx::Conv_655 : Float(576, 96, 1, 1, strides=[96, 1, 1, 1], requires_grad=0, device=cuda:0) = onnx::Identity(%onnx::Conv_646)
  %onnx::Conv_653 : Float(96, strides=[1], requires_grad=0, device=cuda:0) = onnx::Identity(%onnx::Conv_644)
  %onnx::Conv_640 : Float(384, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0) = onnx::Identity(%onnx::Conv_613)
  %onnx::Conv_637 : Float(384, 64, 1, 1, strides=[64, 1, 1, 1], requires_grad=0, device=cuda:0) = onnx::Identity(%onnx::Conv_610)
  %onnx::Conv_634 : Float(64, 384, 1, 1, strides=[384, 1, 1, 1], requires_grad=0, device=cuda:0) = onnx::Identity(%onnx::Conv_616)
  %onnx::Conv_631 : Float(384, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0) = onnx::Identity(%onnx::Conv_613)
  %onnx::Conv_628 : Float(384, 64, 1, 1, strides=[64, 1, 1, 1], requires_grad=0, device=cuda:0) = onnx::Identity(%onnx::Conv_610)
  %onnx::Conv_626 : Float(64, strides=[1], requires_grad=0, device=cuda:0) = onnx::Identity(%onnx::Conv_617)
  %onnx::Conv_625 : Float(64, 384, 1, 1, strides=[384, 1, 1, 1], requires_grad=0, device=cuda:0) = onnx::Identity(%onnx::Conv_616)
  %onnx::Conv_622 : Float(384, 1, 3, 3, strides=[9, 9, 3, 1], requires_grad=0, device=cuda:0) = onnx::Identity(%onnx::Conv_613)
  %onnx::Conv_619 : Float(384, 64, 1, 1, strides=[64, 1, 1, 1], requires_grad=0, device=cuda:0) = onnx::Identity(%onnx::Conv_610)
  %/conv1/conv1.0/conv1.0.0/Conv_output_0 : Float(1, 32, 512, 512, strides=[8388608, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[2, 2], onnx_name="/conv1/conv1.0/conv1.0.0/Conv"](%input.1, %onnx::Conv_547, %onnx::Conv_548), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv1/torch.nn.modules.container.Sequential::conv1.0/torch.nn.modules.conv.Conv2d::conv1.0.0 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv1/conv1.0/conv1.0.2/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/conv1/conv1.0/conv1.0.2/Constant"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv1/torch.nn.modules.container.Sequential::conv1.0/torch.nn.modules.activation.ReLU6::conv1.0.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv1/conv1.0/conv1.0.2/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/conv1/conv1.0/conv1.0.2/Constant_1"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv1/torch.nn.modules.container.Sequential::conv1.0/torch.nn.modules.activation.ReLU6::conv1.0.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv1/conv1.0/conv1.0.2/Clip_output_0 : Float(1, 32, 512, 512, strides=[8388608, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/conv1/conv1.0/conv1.0.2/Clip"](%/conv1/conv1.0/conv1.0.0/Conv_output_0, %/conv1/conv1.0/conv1.0.2/Constant_output_0, %/conv1/conv1.0/conv1.0.2/Constant_1_output_0), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv1/torch.nn.modules.container.Sequential::conv1.0/torch.nn.modules.activation.ReLU6::conv1.0.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv1/conv1.1/conv/conv.0/Conv_output_0 : Float(1, 32, 512, 512, strides=[8388608, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=32, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/conv1/conv1.1/conv/conv.0/Conv"](%/conv1/conv1.0/conv1.0.2/Clip_output_0, %onnx::Conv_550, %onnx::Conv_551), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv1/m2unet_flt.InvertedResidual::conv1.1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv1/conv1.1/conv/conv.2/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/conv1/conv1.1/conv/conv.2/Constant"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv1/m2unet_flt.InvertedResidual::conv1.1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv1/conv1.1/conv/conv.2/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/conv1/conv1.1/conv/conv.2/Constant_1"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv1/m2unet_flt.InvertedResidual::conv1.1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv1/conv1.1/conv/conv.2/Clip_output_0 : Float(1, 32, 512, 512, strides=[8388608, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/conv1/conv1.1/conv/conv.2/Clip"](%/conv1/conv1.1/conv/conv.0/Conv_output_0, %/conv1/conv1.1/conv/conv.2/Constant_output_0, %/conv1/conv1.1/conv/conv.2/Constant_1_output_0), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv1/m2unet_flt.InvertedResidual::conv1.1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv1/conv1.1/conv/conv.3/Conv_output_0 : Float(1, 16, 512, 512, strides=[4194304, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv1/conv1.1/conv/conv.3/Conv"](%/conv1/conv1.1/conv/conv.2/Clip_output_0, %onnx::Conv_553, %onnx::Conv_554), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv1/m2unet_flt.InvertedResidual::conv1.1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv2/conv2.2/conv/conv.0/Conv_output_0 : Float(1, 96, 512, 512, strides=[25165824, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv2/conv2.2/conv/conv.0/Conv"](%/conv1/conv1.1/conv/conv.3/Conv_output_0, %onnx::Conv_556, %onnx::Conv_557), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet_flt.InvertedResidual::conv2.2/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv2/conv2.2/conv/conv.2/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/conv2/conv2.2/conv/conv.2/Constant"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet_flt.InvertedResidual::conv2.2/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv2/conv2.2/conv/conv.2/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/conv2/conv2.2/conv/conv.2/Constant_1"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet_flt.InvertedResidual::conv2.2/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv2/conv2.2/conv/conv.2/Clip_output_0 : Float(1, 96, 512, 512, strides=[25165824, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/conv2/conv2.2/conv/conv.2/Clip"](%/conv2/conv2.2/conv/conv.0/Conv_output_0, %/conv2/conv2.2/conv/conv.2/Constant_output_0, %/conv2/conv2.2/conv/conv.2/Constant_1_output_0), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet_flt.InvertedResidual::conv2.2/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv2/conv2.2/conv/conv.3/Conv_output_0 : Float(1, 96, 256, 256, strides=[6291456, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=96, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[2, 2], onnx_name="/conv2/conv2.2/conv/conv.3/Conv"](%/conv2/conv2.2/conv/conv.2/Clip_output_0, %onnx::Conv_559, %onnx::Conv_560), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet_flt.InvertedResidual::conv2.2/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv2/conv2.2/conv/conv.5/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/conv2/conv2.2/conv/conv.5/Constant"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet_flt.InvertedResidual::conv2.2/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv2/conv2.2/conv/conv.5/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/conv2/conv2.2/conv/conv.5/Constant_1"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet_flt.InvertedResidual::conv2.2/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv2/conv2.2/conv/conv.5/Clip_output_0 : Float(1, 96, 256, 256, strides=[6291456, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/conv2/conv2.2/conv/conv.5/Clip"](%/conv2/conv2.2/conv/conv.3/Conv_output_0, %/conv2/conv2.2/conv/conv.5/Constant_output_0, %/conv2/conv2.2/conv/conv.5/Constant_1_output_0), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet_flt.InvertedResidual::conv2.2/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv2/conv2.2/conv/conv.6/Conv_output_0 : Float(1, 24, 256, 256, strides=[1572864, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv2/conv2.2/conv/conv.6/Conv"](%/conv2/conv2.2/conv/conv.5/Clip_output_0, %onnx::Conv_562, %onnx::Conv_563), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet_flt.InvertedResidual::conv2.2/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv2/conv2.3/conv/conv.0/Conv_output_0 : Float(1, 144, 256, 256, strides=[9437184, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv2/conv2.3/conv/conv.0/Conv"](%/conv2/conv2.2/conv/conv.6/Conv_output_0, %onnx::Conv_565, %onnx::Conv_566), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet_flt.InvertedResidual::conv2.3/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv2/conv2.3/conv/conv.2/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/conv2/conv2.3/conv/conv.2/Constant"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet_flt.InvertedResidual::conv2.3/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv2/conv2.3/conv/conv.2/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/conv2/conv2.3/conv/conv.2/Constant_1"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet_flt.InvertedResidual::conv2.3/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv2/conv2.3/conv/conv.2/Clip_output_0 : Float(1, 144, 256, 256, strides=[9437184, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/conv2/conv2.3/conv/conv.2/Clip"](%/conv2/conv2.3/conv/conv.0/Conv_output_0, %/conv2/conv2.3/conv/conv.2/Constant_output_0, %/conv2/conv2.3/conv/conv.2/Constant_1_output_0), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet_flt.InvertedResidual::conv2.3/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv2/conv2.3/conv/conv.3/Conv_output_0 : Float(1, 144, 256, 256, strides=[9437184, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=144, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/conv2/conv2.3/conv/conv.3/Conv"](%/conv2/conv2.3/conv/conv.2/Clip_output_0, %onnx::Conv_568, %onnx::Conv_569), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet_flt.InvertedResidual::conv2.3/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv2/conv2.3/conv/conv.5/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/conv2/conv2.3/conv/conv.5/Constant"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet_flt.InvertedResidual::conv2.3/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv2/conv2.3/conv/conv.5/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/conv2/conv2.3/conv/conv.5/Constant_1"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet_flt.InvertedResidual::conv2.3/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv2/conv2.3/conv/conv.5/Clip_output_0 : Float(1, 144, 256, 256, strides=[9437184, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/conv2/conv2.3/conv/conv.5/Clip"](%/conv2/conv2.3/conv/conv.3/Conv_output_0, %/conv2/conv2.3/conv/conv.5/Constant_output_0, %/conv2/conv2.3/conv/conv.5/Constant_1_output_0), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet_flt.InvertedResidual::conv2.3/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv2/conv2.3/conv/conv.6/Conv_output_0 : Float(1, 24, 256, 256, strides=[1572864, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv2/conv2.3/conv/conv.6/Conv"](%/conv2/conv2.3/conv/conv.5/Clip_output_0, %onnx::Conv_571, %onnx::Conv_572), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet_flt.InvertedResidual::conv2.3/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv2/conv2.3/Add_output_0 : Float(1, 24, 256, 256, strides=[1572864, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Add[onnx_name="/conv2/conv2.3/Add"](%/conv2/conv2.2/conv/conv.6/Conv_output_0, %/conv2/conv2.3/conv/conv.6/Conv_output_0), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv2/m2unet_flt.InvertedResidual::conv2.3 # /media/prakashlab/SSD/quantize_m2unet/m2unet_flt.py:85:0
  %/conv3/conv3.4/conv/conv.0/Conv_output_0 : Float(1, 144, 256, 256, strides=[9437184, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv3/conv3.4/conv/conv.0/Conv"](%/conv2/conv2.3/Add_output_0, %onnx::Conv_574, %onnx::Conv_575), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.4/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv3/conv3.4/conv/conv.2/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/conv3/conv3.4/conv/conv.2/Constant"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.4/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv3/conv3.4/conv/conv.2/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/conv3/conv3.4/conv/conv.2/Constant_1"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.4/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv3/conv3.4/conv/conv.2/Clip_output_0 : Float(1, 144, 256, 256, strides=[9437184, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/conv3/conv3.4/conv/conv.2/Clip"](%/conv3/conv3.4/conv/conv.0/Conv_output_0, %/conv3/conv3.4/conv/conv.2/Constant_output_0, %/conv3/conv3.4/conv/conv.2/Constant_1_output_0), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.4/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv3/conv3.4/conv/conv.3/Conv_output_0 : Float(1, 144, 128, 128, strides=[2359296, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=144, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[2, 2], onnx_name="/conv3/conv3.4/conv/conv.3/Conv"](%/conv3/conv3.4/conv/conv.2/Clip_output_0, %onnx::Conv_577, %onnx::Conv_578), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.4/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv3/conv3.4/conv/conv.5/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/conv3/conv3.4/conv/conv.5/Constant"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.4/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv3/conv3.4/conv/conv.5/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/conv3/conv3.4/conv/conv.5/Constant_1"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.4/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv3/conv3.4/conv/conv.5/Clip_output_0 : Float(1, 144, 128, 128, strides=[2359296, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/conv3/conv3.4/conv/conv.5/Clip"](%/conv3/conv3.4/conv/conv.3/Conv_output_0, %/conv3/conv3.4/conv/conv.5/Constant_output_0, %/conv3/conv3.4/conv/conv.5/Constant_1_output_0), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.4/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv3/conv3.4/conv/conv.6/Conv_output_0 : Float(1, 32, 128, 128, strides=[524288, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv3/conv3.4/conv/conv.6/Conv"](%/conv3/conv3.4/conv/conv.5/Clip_output_0, %onnx::Conv_580, %onnx::Conv_581), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.4/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv3/conv3.5/conv/conv.0/Conv_output_0 : Float(1, 192, 128, 128, strides=[3145728, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv3/conv3.5/conv/conv.0/Conv"](%/conv3/conv3.4/conv/conv.6/Conv_output_0, %onnx::Conv_583, %onnx::Conv_584), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.5/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv3/conv3.5/conv/conv.2/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/conv3/conv3.5/conv/conv.2/Constant"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.5/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv3/conv3.5/conv/conv.2/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/conv3/conv3.5/conv/conv.2/Constant_1"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.5/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv3/conv3.5/conv/conv.2/Clip_output_0 : Float(1, 192, 128, 128, strides=[3145728, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/conv3/conv3.5/conv/conv.2/Clip"](%/conv3/conv3.5/conv/conv.0/Conv_output_0, %/conv3/conv3.5/conv/conv.2/Constant_output_0, %/conv3/conv3.5/conv/conv.2/Constant_1_output_0), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.5/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv3/conv3.5/conv/conv.3/Conv_output_0 : Float(1, 192, 128, 128, strides=[3145728, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=192, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/conv3/conv3.5/conv/conv.3/Conv"](%/conv3/conv3.5/conv/conv.2/Clip_output_0, %onnx::Conv_586, %onnx::Conv_587), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.5/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv3/conv3.5/conv/conv.5/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/conv3/conv3.5/conv/conv.5/Constant"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.5/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv3/conv3.5/conv/conv.5/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/conv3/conv3.5/conv/conv.5/Constant_1"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.5/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv3/conv3.5/conv/conv.5/Clip_output_0 : Float(1, 192, 128, 128, strides=[3145728, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/conv3/conv3.5/conv/conv.5/Clip"](%/conv3/conv3.5/conv/conv.3/Conv_output_0, %/conv3/conv3.5/conv/conv.5/Constant_output_0, %/conv3/conv3.5/conv/conv.5/Constant_1_output_0), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.5/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv3/conv3.5/conv/conv.6/Conv_output_0 : Float(1, 32, 128, 128, strides=[524288, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv3/conv3.5/conv/conv.6/Conv"](%/conv3/conv3.5/conv/conv.5/Clip_output_0, %onnx::Conv_589, %onnx::Conv_590), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.5/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv3/conv3.5/Add_output_0 : Float(1, 32, 128, 128, strides=[524288, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Add[onnx_name="/conv3/conv3.5/Add"](%/conv3/conv3.4/conv/conv.6/Conv_output_0, %/conv3/conv3.5/conv/conv.6/Conv_output_0), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.5 # /media/prakashlab/SSD/quantize_m2unet/m2unet_flt.py:85:0
  %/conv3/conv3.6/conv/conv.0/Conv_output_0 : Float(1, 192, 128, 128, strides=[3145728, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv3/conv3.6/conv/conv.0/Conv"](%/conv3/conv3.5/Add_output_0, %onnx::Conv_592, %onnx::Conv_593), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.6/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv3/conv3.6/conv/conv.2/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/conv3/conv3.6/conv/conv.2/Constant"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.6/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv3/conv3.6/conv/conv.2/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/conv3/conv3.6/conv/conv.2/Constant_1"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.6/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv3/conv3.6/conv/conv.2/Clip_output_0 : Float(1, 192, 128, 128, strides=[3145728, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/conv3/conv3.6/conv/conv.2/Clip"](%/conv3/conv3.6/conv/conv.0/Conv_output_0, %/conv3/conv3.6/conv/conv.2/Constant_output_0, %/conv3/conv3.6/conv/conv.2/Constant_1_output_0), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.6/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv3/conv3.6/conv/conv.3/Conv_output_0 : Float(1, 192, 128, 128, strides=[3145728, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=192, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/conv3/conv3.6/conv/conv.3/Conv"](%/conv3/conv3.6/conv/conv.2/Clip_output_0, %onnx::Conv_595, %onnx::Conv_596), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.6/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv3/conv3.6/conv/conv.5/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/conv3/conv3.6/conv/conv.5/Constant"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.6/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv3/conv3.6/conv/conv.5/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/conv3/conv3.6/conv/conv.5/Constant_1"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.6/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv3/conv3.6/conv/conv.5/Clip_output_0 : Float(1, 192, 128, 128, strides=[3145728, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/conv3/conv3.6/conv/conv.5/Clip"](%/conv3/conv3.6/conv/conv.3/Conv_output_0, %/conv3/conv3.6/conv/conv.5/Constant_output_0, %/conv3/conv3.6/conv/conv.5/Constant_1_output_0), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.6/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv3/conv3.6/conv/conv.6/Conv_output_0 : Float(1, 32, 128, 128, strides=[524288, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv3/conv3.6/conv/conv.6/Conv"](%/conv3/conv3.6/conv/conv.5/Clip_output_0, %onnx::Conv_598, %onnx::Conv_599), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.6/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv3/conv3.6/Add_output_0 : Float(1, 32, 128, 128, strides=[524288, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Add[onnx_name="/conv3/conv3.6/Add"](%/conv3/conv3.5/Add_output_0, %/conv3/conv3.6/conv/conv.6/Conv_output_0), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv3/m2unet_flt.InvertedResidual::conv3.6 # /media/prakashlab/SSD/quantize_m2unet/m2unet_flt.py:85:0
  %/conv4/conv4.12/Add_output_0 : Float(1, 96, 64, 64, strides=[393216, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Add[onnx_name="/conv4/conv4.12/Add"](%/conv4/conv4.11/conv/conv.6/Conv_output_0, %/conv4/conv4.12/conv/conv.6/Conv_output_0), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet_flt.InvertedResidual::conv4.12 # /media/prakashlab/SSD/quantize_m2unet/m2unet_flt.py:85:0
  %/conv4/conv4.13/conv/conv.0/Conv_output_0 : Float(1, 576, 64, 64, strides=[2359296, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv4/conv4.13/conv/conv.0/Conv"](%/conv4/conv4.12/Add_output_0, %onnx::Conv_655, %onnx::Conv_656), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet_flt.InvertedResidual::conv4.13/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv4/conv4.13/conv/conv.2/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/conv4/conv4.13/conv/conv.2/Constant"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet_flt.InvertedResidual::conv4.13/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv4/conv4.13/conv/conv.2/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/conv4/conv4.13/conv/conv.2/Constant_1"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet_flt.InvertedResidual::conv4.13/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv4/conv4.13/conv/conv.2/Clip_output_0 : Float(1, 576, 64, 64, strides=[2359296, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/conv4/conv4.13/conv/conv.2/Clip"](%/conv4/conv4.13/conv/conv.0/Conv_output_0, %/conv4/conv4.13/conv/conv.2/Constant_output_0, %/conv4/conv4.13/conv/conv.2/Constant_1_output_0), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet_flt.InvertedResidual::conv4.13/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv4/conv4.13/conv/conv.3/Conv_output_0 : Float(1, 576, 64, 64, strides=[2359296, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=576, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/conv4/conv4.13/conv/conv.3/Conv"](%/conv4/conv4.13/conv/conv.2/Clip_output_0, %onnx::Conv_658, %onnx::Conv_659), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet_flt.InvertedResidual::conv4.13/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv4/conv4.13/conv/conv.5/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/conv4/conv4.13/conv/conv.5/Constant"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet_flt.InvertedResidual::conv4.13/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv4/conv4.13/conv/conv.5/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/conv4/conv4.13/conv/conv.5/Constant_1"](), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet_flt.InvertedResidual::conv4.13/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv4/conv4.13/conv/conv.5/Clip_output_0 : Float(1, 576, 64, 64, strides=[2359296, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/conv4/conv4.13/conv/conv.5/Clip"](%/conv4/conv4.13/conv/conv.3/Conv_output_0, %/conv4/conv4.13/conv/conv.5/Constant_output_0, %/conv4/conv4.13/conv/conv.5/Constant_1_output_0), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet_flt.InvertedResidual::conv4.13/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/conv4/conv4.13/conv/conv.6/Conv_output_0 : Float(1, 96, 64, 64, strides=[393216, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/conv4/conv4.13/conv/conv.6/Conv"](%/conv4/conv4.13/conv/conv.5/Clip_output_0, %onnx::Conv_661, %onnx::Conv_662), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet_flt.InvertedResidual::conv4.13/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/conv4/conv4.13/Add_output_0 : Float(1, 96, 64, 64, strides=[393216, 4096, 64, 1], requires_grad=1, device=cuda:0) = onnx::Add[onnx_name="/conv4/conv4.13/Add"](%/conv4/conv4.12/Add_output_0, %/conv4/conv4.13/conv/conv.6/Conv_output_0), scope: m2unet_flt.M2UNet::/torch.nn.modules.container.Sequential::conv4/m2unet_flt.InvertedResidual::conv4.13 # /media/prakashlab/SSD/quantize_m2unet/m2unet_flt.py:85:0
  %/decode4/upsample/Constant_output_0 : Float(4, strides=[1], requires_grad=0, device=cpu) = onnx::Constant[value= 1  1  2  2 [ CPUFloatType{4} ], onnx_name="/decode4/upsample/Constant"](), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode4/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:3946:0
  %/decode4/upsample/Constant_1_output_0 : Float(0, strides=[1], device=cpu) = onnx::Constant[value=[ CPUFloatType{0} ], onnx_name="/decode4/upsample/Constant_1"](), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode4/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:3946:0
  %/decode4/upsample/Resize_output_0 : Float(1, 96, 128, 128, strides=[1572864, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Resize[coordinate_transformation_mode="half_pixel", cubic_coeff_a=-0.75, mode="linear", nearest_mode="floor", onnx_name="/decode4/upsample/Resize"](%/conv4/conv4.13/Add_output_0, %/decode4/upsample/Constant_1_output_0, %/decode4/upsample/Constant_output_0), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode4/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:3946:0
  %/decode4/Concat_output_0 : Float(1, 128, 128, 128, strides=[2097152, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Concat[axis=1, onnx_name="/decode4/Concat"](%/decode4/upsample/Resize_output_0, %/conv3/conv3.6/Add_output_0), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode4 # /media/prakashlab/SSD/quantize_m2unet/m2unet_flt.py:130:0
  %/decode4/ir1/conv/conv.0/Conv_output_0 : Float(1, 19, 128, 128, strides=[311296, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/decode4/ir1/conv/conv.0/Conv"](%/decode4/Concat_output_0, %onnx::Conv_664, %onnx::Conv_665), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode4/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/decode4/ir1/conv/conv.2/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/decode4/ir1/conv/conv.2/Constant"](), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode4/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode4/ir1/conv/conv.2/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/decode4/ir1/conv/conv.2/Constant_1"](), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode4/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode4/ir1/conv/conv.2/Clip_output_0 : Float(1, 19, 128, 128, strides=[311296, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/decode4/ir1/conv/conv.2/Clip"](%/decode4/ir1/conv/conv.0/Conv_output_0, %/decode4/ir1/conv/conv.2/Constant_output_0, %/decode4/ir1/conv/conv.2/Constant_1_output_0), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode4/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode4/ir1/conv/conv.3/Conv_output_0 : Float(1, 19, 128, 128, strides=[311296, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=19, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/decode4/ir1/conv/conv.3/Conv"](%/decode4/ir1/conv/conv.2/Clip_output_0, %onnx::Conv_667, %onnx::Conv_668), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode4/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/decode4/ir1/conv/conv.5/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/decode4/ir1/conv/conv.5/Constant"](), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode4/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode4/ir1/conv/conv.5/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/decode4/ir1/conv/conv.5/Constant_1"](), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode4/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode4/ir1/conv/conv.5/Clip_output_0 : Float(1, 19, 128, 128, strides=[311296, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/decode4/ir1/conv/conv.5/Clip"](%/decode4/ir1/conv/conv.3/Conv_output_0, %/decode4/ir1/conv/conv.5/Constant_output_0, %/decode4/ir1/conv/conv.5/Constant_1_output_0), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode4/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode4/ir1/conv/conv.6/Conv_output_0 : Float(1, 64, 128, 128, strides=[1048576, 16384, 128, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/decode4/ir1/conv/conv.6/Conv"](%/decode4/ir1/conv/conv.5/Clip_output_0, %onnx::Conv_670, %onnx::Conv_671), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode4/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/decode3/upsample/Constant_output_0 : Float(4, strides=[1], requires_grad=0, device=cpu) = onnx::Constant[value= 1  1  2  2 [ CPUFloatType{4} ], onnx_name="/decode3/upsample/Constant"](), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode3/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:3946:0
  %/decode3/upsample/Constant_1_output_0 : Float(0, strides=[1], device=cpu) = onnx::Constant[value=[ CPUFloatType{0} ], onnx_name="/decode3/upsample/Constant_1"](), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode3/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:3946:0
  %/decode3/upsample/Resize_output_0 : Float(1, 64, 256, 256, strides=[4194304, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Resize[coordinate_transformation_mode="half_pixel", cubic_coeff_a=-0.75, mode="linear", nearest_mode="floor", onnx_name="/decode3/upsample/Resize"](%/decode4/ir1/conv/conv.6/Conv_output_0, %/decode3/upsample/Constant_1_output_0, %/decode3/upsample/Constant_output_0), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode3/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:3946:0
  %/decode3/Concat_output_0 : Float(1, 88, 256, 256, strides=[5767168, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Concat[axis=1, onnx_name="/decode3/Concat"](%/decode3/upsample/Resize_output_0, %/conv2/conv2.3/Add_output_0), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode3 # /media/prakashlab/SSD/quantize_m2unet/m2unet_flt.py:130:0
  %/decode3/ir1/conv/conv.0/Conv_output_0 : Float(1, 13, 256, 256, strides=[851968, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/decode3/ir1/conv/conv.0/Conv"](%/decode3/Concat_output_0, %onnx::Conv_673, %onnx::Conv_674), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode3/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/decode3/ir1/conv/conv.2/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/decode3/ir1/conv/conv.2/Constant"](), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode3/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode3/ir1/conv/conv.2/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/decode3/ir1/conv/conv.2/Constant_1"](), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode3/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode3/ir1/conv/conv.2/Clip_output_0 : Float(1, 13, 256, 256, strides=[851968, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/decode3/ir1/conv/conv.2/Clip"](%/decode3/ir1/conv/conv.0/Conv_output_0, %/decode3/ir1/conv/conv.2/Constant_output_0, %/decode3/ir1/conv/conv.2/Constant_1_output_0), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode3/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode3/ir1/conv/conv.3/Conv_output_0 : Float(1, 13, 256, 256, strides=[851968, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=13, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/decode3/ir1/conv/conv.3/Conv"](%/decode3/ir1/conv/conv.2/Clip_output_0, %onnx::Conv_676, %onnx::Conv_677), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode3/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/decode3/ir1/conv/conv.5/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/decode3/ir1/conv/conv.5/Constant"](), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode3/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode3/ir1/conv/conv.5/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/decode3/ir1/conv/conv.5/Constant_1"](), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode3/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode3/ir1/conv/conv.5/Clip_output_0 : Float(1, 13, 256, 256, strides=[851968, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/decode3/ir1/conv/conv.5/Clip"](%/decode3/ir1/conv/conv.3/Conv_output_0, %/decode3/ir1/conv/conv.5/Constant_output_0, %/decode3/ir1/conv/conv.5/Constant_1_output_0), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode3/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode3/ir1/conv/conv.6/Conv_output_0 : Float(1, 44, 256, 256, strides=[2883584, 65536, 256, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/decode3/ir1/conv/conv.6/Conv"](%/decode3/ir1/conv/conv.5/Clip_output_0, %onnx::Conv_679, %onnx::Conv_680), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode3/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/decode2/upsample/Constant_output_0 : Float(4, strides=[1], requires_grad=0, device=cpu) = onnx::Constant[value= 1  1  2  2 [ CPUFloatType{4} ], onnx_name="/decode2/upsample/Constant"](), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode2/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:3946:0
  %/decode2/upsample/Constant_1_output_0 : Float(0, strides=[1], device=cpu) = onnx::Constant[value=[ CPUFloatType{0} ], onnx_name="/decode2/upsample/Constant_1"](), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode2/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:3946:0
  %/decode2/upsample/Resize_output_0 : Float(1, 44, 512, 512, strides=[11534336, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Resize[coordinate_transformation_mode="half_pixel", cubic_coeff_a=-0.75, mode="linear", nearest_mode="floor", onnx_name="/decode2/upsample/Resize"](%/decode3/ir1/conv/conv.6/Conv_output_0, %/decode2/upsample/Constant_1_output_0, %/decode2/upsample/Constant_output_0), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode2/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:3946:0
  %/decode2/Concat_output_0 : Float(1, 60, 512, 512, strides=[15728640, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Concat[axis=1, onnx_name="/decode2/Concat"](%/decode2/upsample/Resize_output_0, %/conv1/conv1.1/conv/conv.3/Conv_output_0), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode2 # /media/prakashlab/SSD/quantize_m2unet/m2unet_flt.py:130:0
  %/decode2/ir1/conv/conv.0/Conv_output_0 : Float(1, 9, 512, 512, strides=[2359296, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/decode2/ir1/conv/conv.0/Conv"](%/decode2/Concat_output_0, %onnx::Conv_682, %onnx::Conv_683), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode2/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/decode2/ir1/conv/conv.2/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/decode2/ir1/conv/conv.2/Constant"](), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode2/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode2/ir1/conv/conv.2/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/decode2/ir1/conv/conv.2/Constant_1"](), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode2/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode2/ir1/conv/conv.2/Clip_output_0 : Float(1, 9, 512, 512, strides=[2359296, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/decode2/ir1/conv/conv.2/Clip"](%/decode2/ir1/conv/conv.0/Conv_output_0, %/decode2/ir1/conv/conv.2/Constant_output_0, %/decode2/ir1/conv/conv.2/Constant_1_output_0), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode2/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode2/ir1/conv/conv.3/Conv_output_0 : Float(1, 9, 512, 512, strides=[2359296, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=9, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/decode2/ir1/conv/conv.3/Conv"](%/decode2/ir1/conv/conv.2/Clip_output_0, %onnx::Conv_685, %onnx::Conv_686), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode2/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/decode2/ir1/conv/conv.5/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/decode2/ir1/conv/conv.5/Constant"](), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode2/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode2/ir1/conv/conv.5/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/decode2/ir1/conv/conv.5/Constant_1"](), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode2/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode2/ir1/conv/conv.5/Clip_output_0 : Float(1, 9, 512, 512, strides=[2359296, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/decode2/ir1/conv/conv.5/Clip"](%/decode2/ir1/conv/conv.3/Conv_output_0, %/decode2/ir1/conv/conv.5/Constant_output_0, %/decode2/ir1/conv/conv.5/Constant_1_output_0), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode2/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode2/ir1/conv/conv.6/Conv_output_0 : Float(1, 30, 512, 512, strides=[7864320, 262144, 512, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/decode2/ir1/conv/conv.6/Conv"](%/decode2/ir1/conv/conv.5/Clip_output_0, %onnx::Conv_688, %onnx::Conv_689), scope: m2unet_flt.M2UNet::/m2unet_flt.DecoderBlock::decode2/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/decode1/upsample/Constant_output_0 : Float(4, strides=[1], requires_grad=0, device=cpu) = onnx::Constant[value= 1  1  2  2 [ CPUFloatType{4} ], onnx_name="/decode1/upsample/Constant"](), scope: m2unet_flt.M2UNet::/m2unet_flt.LastDecoderBlock::decode1/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:3946:0
  %/decode1/upsample/Constant_1_output_0 : Float(0, strides=[1], device=cpu) = onnx::Constant[value=[ CPUFloatType{0} ], onnx_name="/decode1/upsample/Constant_1"](), scope: m2unet_flt.M2UNet::/m2unet_flt.LastDecoderBlock::decode1/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:3946:0
  %/decode1/upsample/Resize_output_0 : Float(1, 30, 1024, 1024, strides=[31457280, 1048576, 1024, 1], requires_grad=1, device=cuda:0) = onnx::Resize[coordinate_transformation_mode="half_pixel", cubic_coeff_a=-0.75, mode="linear", nearest_mode="floor", onnx_name="/decode1/upsample/Resize"](%/decode2/ir1/conv/conv.6/Conv_output_0, %/decode1/upsample/Constant_1_output_0, %/decode1/upsample/Constant_output_0), scope: m2unet_flt.M2UNet::/m2unet_flt.LastDecoderBlock::decode1/torch.nn.modules.upsampling.Upsample::upsample # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:3946:0
  %/decode1/Concat_output_0 : Float(1, 31, 1024, 1024, strides=[32505856, 1048576, 1024, 1], requires_grad=1, device=cuda:0) = onnx::Concat[axis=1, onnx_name="/decode1/Concat"](%/decode1/upsample/Resize_output_0, %input.1), scope: m2unet_flt.M2UNet::/m2unet_flt.LastDecoderBlock::decode1 # /media/prakashlab/SSD/quantize_m2unet/m2unet_flt.py:156:0
  %/decode1/ir1/conv/conv.0/Conv_output_0 : Float(1, 5, 1024, 1024, strides=[5242880, 1048576, 1024, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/decode1/ir1/conv/conv.0/Conv"](%/decode1/Concat_output_0, %onnx::Conv_691, %onnx::Conv_692), scope: m2unet_flt.M2UNet::/m2unet_flt.LastDecoderBlock::decode1/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/decode1/ir1/conv/conv.2/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/decode1/ir1/conv/conv.2/Constant"](), scope: m2unet_flt.M2UNet::/m2unet_flt.LastDecoderBlock::decode1/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode1/ir1/conv/conv.2/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/decode1/ir1/conv/conv.2/Constant_1"](), scope: m2unet_flt.M2UNet::/m2unet_flt.LastDecoderBlock::decode1/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode1/ir1/conv/conv.2/Clip_output_0 : Float(1, 5, 1024, 1024, strides=[5242880, 1048576, 1024, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/decode1/ir1/conv/conv.2/Clip"](%/decode1/ir1/conv/conv.0/Conv_output_0, %/decode1/ir1/conv/conv.2/Constant_output_0, %/decode1/ir1/conv/conv.2/Constant_1_output_0), scope: m2unet_flt.M2UNet::/m2unet_flt.LastDecoderBlock::decode1/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.2 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode1/ir1/conv/conv.3/Conv_output_0 : Float(1, 5, 1024, 1024, strides=[5242880, 1048576, 1024, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=5, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1], onnx_name="/decode1/ir1/conv/conv.3/Conv"](%/decode1/ir1/conv/conv.2/Clip_output_0, %onnx::Conv_694, %onnx::Conv_695), scope: m2unet_flt.M2UNet::/m2unet_flt.LastDecoderBlock::decode1/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.3 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %/decode1/ir1/conv/conv.5/Constant_output_0 : Float(device=cpu) = onnx::Constant[value={0}, onnx_name="/decode1/ir1/conv/conv.5/Constant"](), scope: m2unet_flt.M2UNet::/m2unet_flt.LastDecoderBlock::decode1/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode1/ir1/conv/conv.5/Constant_1_output_0 : Float(device=cpu) = onnx::Constant[value={6}, onnx_name="/decode1/ir1/conv/conv.5/Constant_1"](), scope: m2unet_flt.M2UNet::/m2unet_flt.LastDecoderBlock::decode1/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode1/ir1/conv/conv.5/Clip_output_0 : Float(1, 5, 1024, 1024, strides=[5242880, 1048576, 1024, 1], requires_grad=1, device=cuda:0) = onnx::Clip[onnx_name="/decode1/ir1/conv/conv.5/Clip"](%/decode1/ir1/conv/conv.3/Conv_output_0, %/decode1/ir1/conv/conv.5/Constant_output_0, %/decode1/ir1/conv/conv.5/Constant_1_output_0), scope: m2unet_flt.M2UNet::/m2unet_flt.LastDecoderBlock::decode1/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.activation.ReLU6::conv.5 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/functional.py:1506:0
  %/decode1/ir1/conv/conv.6/Conv_output_0 : Float(1, 16, 1024, 1024, strides=[16777216, 1048576, 1024, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/decode1/ir1/conv/conv.6/Conv"](%/decode1/ir1/conv/conv.5/Clip_output_0, %onnx::Conv_697, %onnx::Conv_698), scope: m2unet_flt.M2UNet::/m2unet_flt.LastDecoderBlock::decode1/m2unet_flt.InvertedResidual::ir1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.6 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  %545 : Float(1, 1, 1024, 1024, strides=[1048576, 1048576, 1024, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1], onnx_name="/decode1/conv/conv.0/Conv"](%/decode1/ir1/conv/conv.6/Conv_output_0, %decode1.conv.0.weight, %decode1.conv.0.bias), scope: m2unet_flt.M2UNet::/m2unet_flt.LastDecoderBlock::decode1/torch.nn.modules.container.Sequential::conv/torch.nn.modules.conv.Conv2d::conv.0 # /home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py:459:0
  return (%545)

Here’s the log when running trtexec:

&&&& RUNNING TensorRT.trtexec [TensorRT v8401] # ./trtexec --onnx=/media/prakashlab/SSD/quantize_m2unet/models/flat_model_24_9_sz1024_v11.onnx --saveEngine=/media/prakashlab/SSD/quantize_m2unet/models/flat_model_24_9_sz1024_v11_from_onnx.trt --explicitBatch
[03/23/2023-10:23:29] [W] --explicitBatch flag has been deprecated and has no effect!
[03/23/2023-10:23:29] [W] Explicit batch dim is automatically enabled if input model is ONNX or if dynamic shapes are provided when the engine is built.
[03/23/2023-10:23:29] [I] === Model Options ===
[03/23/2023-10:23:29] [I] Format: ONNX
[03/23/2023-10:23:29] [I] Model: /media/prakashlab/SSD/quantize_m2unet/models/flat_model_24_9_sz1024_v11.onnx
[03/23/2023-10:23:29] [I] Output:
[03/23/2023-10:23:29] [I] === Build Options ===
[03/23/2023-10:23:29] [I] Max batch: explicit batch
[03/23/2023-10:23:29] [I] Memory Pools: workspace: default, dlaSRAM: default, dlaLocalDRAM: default, dlaGlobalDRAM: default
[03/23/2023-10:23:29] [I] minTiming: 1
[03/23/2023-10:23:29] [I] avgTiming: 8
[03/23/2023-10:23:29] [I] Precision: FP32
[03/23/2023-10:23:29] [I] LayerPrecisions: 
[03/23/2023-10:23:29] [I] Calibration: 
[03/23/2023-10:23:29] [I] Refit: Disabled
[03/23/2023-10:23:29] [I] Sparsity: Disabled
[03/23/2023-10:23:29] [I] Safe mode: Disabled
[03/23/2023-10:23:29] [I] DirectIO mode: Disabled
[03/23/2023-10:23:29] [I] Restricted mode: Disabled
[03/23/2023-10:23:29] [I] Build only: Disabled
[03/23/2023-10:23:29] [I] Save engine: /media/prakashlab/SSD/quantize_m2unet/models/flat_model_24_9_sz1024_v11_from_onnx.trt
[03/23/2023-10:23:29] [I] Load engine: 
[03/23/2023-10:23:29] [I] Profiling verbosity: 0
[03/23/2023-10:23:29] [I] Tactic sources: Using default tactic sources
[03/23/2023-10:23:29] [I] timingCacheMode: local
[03/23/2023-10:23:29] [I] timingCacheFile: 
[03/23/2023-10:23:29] [I] Input(s)s format: fp32:CHW
[03/23/2023-10:23:29] [I] Output(s)s format: fp32:CHW
[03/23/2023-10:23:29] [I] Input build shapes: model
[03/23/2023-10:23:29] [I] Input calibration shapes: model
[03/23/2023-10:23:29] [I] === System Options ===
[03/23/2023-10:23:29] [I] Device: 0
[03/23/2023-10:23:29] [I] DLACore: 
[03/23/2023-10:23:29] [I] Plugins:
[03/23/2023-10:23:29] [I] === Inference Options ===
[03/23/2023-10:23:29] [I] Batch: Explicit
[03/23/2023-10:23:29] [I] Input inference shapes: model
[03/23/2023-10:23:29] [I] Iterations: 10
[03/23/2023-10:23:29] [I] Duration: 3s (+ 200ms warm up)
[03/23/2023-10:23:29] [I] Sleep time: 0ms
[03/23/2023-10:23:29] [I] Idle time: 0ms
[03/23/2023-10:23:29] [I] Streams: 1
[03/23/2023-10:23:29] [I] ExposeDMA: Disabled
[03/23/2023-10:23:29] [I] Data transfers: Enabled
[03/23/2023-10:23:29] [I] Spin-wait: Disabled
[03/23/2023-10:23:29] [I] Multithreading: Disabled
[03/23/2023-10:23:29] [I] CUDA Graph: Disabled
[03/23/2023-10:23:29] [I] Separate profiling: Disabled
[03/23/2023-10:23:29] [I] Time Deserialize: Disabled
[03/23/2023-10:23:29] [I] Time Refit: Disabled
[03/23/2023-10:23:29] [I] Inputs:
[03/23/2023-10:23:29] [I] === Reporting Options ===
[03/23/2023-10:23:29] [I] Verbose: Disabled
[03/23/2023-10:23:29] [I] Averages: 10 inferences
[03/23/2023-10:23:29] [I] Percentile: 99
[03/23/2023-10:23:29] [I] Dump refittable layers:Disabled
[03/23/2023-10:23:29] [I] Dump output: Disabled
[03/23/2023-10:23:29] [I] Profile: Disabled
[03/23/2023-10:23:29] [I] Export timing to JSON file: 
[03/23/2023-10:23:29] [I] Export output to JSON file: 
[03/23/2023-10:23:29] [I] Export profile to JSON file: 
[03/23/2023-10:23:29] [I] 
[03/23/2023-10:23:29] [I] === Device Information ===
[03/23/2023-10:23:29] [I] Selected Device: Xavier
[03/23/2023-10:23:29] [I] Compute Capability: 7.2
[03/23/2023-10:23:29] [I] SMs: 6
[03/23/2023-10:23:29] [I] Compute Clock Rate: 1.109 GHz
[03/23/2023-10:23:29] [I] Device Global Memory: 6854 MiB
[03/23/2023-10:23:29] [I] Shared Memory per SM: 96 KiB
[03/23/2023-10:23:29] [I] Memory Bus Width: 256 bits (ECC disabled)
[03/23/2023-10:23:29] [I] Memory Clock Rate: 1.109 GHz
[03/23/2023-10:23:29] [I] 
[03/23/2023-10:23:29] [I] TensorRT version: 8.4.1
[03/23/2023-10:23:39] [I] [TRT] [MemUsageChange] Init CUDA: CPU +185, GPU +0, now: CPU 209, GPU 2882 (MiB)
[03/23/2023-10:23:44] [I] [TRT] [MemUsageChange] Init builder kernel library: CPU +131, GPU +123, now: CPU 359, GPU 3022 (MiB)
[03/23/2023-10:23:44] [I] Start parsing network model
[03/23/2023-10:23:44] [I] [TRT] ----------------------------------------------------------------
[03/23/2023-10:23:44] [I] [TRT] Input filename:   /media/prakashlab/SSD/quantize_m2unet/models/flat_model_24_9_sz1024_v11.onnx
[03/23/2023-10:23:44] [I] [TRT] ONNX IR version:  0.0.6
[03/23/2023-10:23:44] [I] [TRT] Opset version:    11
[03/23/2023-10:23:44] [I] [TRT] Producer name:    pytorch
[03/23/2023-10:23:44] [I] [TRT] Producer version: 1.13.0
[03/23/2023-10:23:44] [I] [TRT] Domain:           
[03/23/2023-10:23:44] [I] [TRT] Model version:    0
[03/23/2023-10:23:44] [I] [TRT] Doc string:       
[03/23/2023-10:23:44] [I] [TRT] ----------------------------------------------------------------
[03/23/2023-10:23:44] [E] Error[4]: /conv4/conv4.9/conv/conv.0/Conv: two inputs (data and weights) are allowed only in explicit-quantization mode.
[03/23/2023-10:23:44] [E] [TRT] ModelImporter.cpp:773: While parsing node number 90 [Conv -> "/conv4/conv4.9/conv/conv.0/Conv_output_0"]:
[03/23/2023-10:23:44] [E] [TRT] ModelImporter.cpp:774: --- Begin node ---
[03/23/2023-10:23:44] [E] [TRT] ModelImporter.cpp:775: input: "/conv4/conv4.8/Add_output_0"
input: "onnx::Conv_619"
input: "onnx::Conv_620"
output: "/conv4/conv4.9/conv/conv.0/Conv_output_0"
name: "/conv4/conv4.9/conv/conv.0/Conv"
op_type: "Conv"
attribute {
  name: "dilations"
  ints: 1
  ints: 1
  type: INTS
}
attribute {
  name: "group"
  i: 1
  type: INT
}
attribute {
  name: "kernel_shape"
  ints: 1
  ints: 1
  type: INTS
}
attribute {
  name: "pads"
  ints: 0
  ints: 0
  ints: 0
  ints: 0
  type: INTS
}
attribute {
  name: "strides"
  ints: 1
  ints: 1
  type: INTS
}
doc_string: "/home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py(459): _conv_forward\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/conv.py(463): forward\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/module.py(1178): _slow_forward\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/module.py(1190): _call_impl\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/container.py(204): forward\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/module.py(1178): _slow_forward\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/module.py(1190): _call_impl\n/media/prakashlab/SSD/quantize_m2unet/m2unet_flt.py(85): forward\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/module.py(1178): _slow_forward\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/module.py(1190): _call_impl\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/container.py(204): forward\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/module.py(1178): _slow_forward\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/module.py(1190): _call_impl\n/media/prakashlab/SSD/quantize_m2unet/m2unet_flt.py(197): forward\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/module.py(1178): _slow_forward\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/module.py(1190): _call_impl\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/jit/_trace.py(118): wrapper\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/jit/_trace.py(127): forward\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/nn/modules/module.py(1190): _call_impl\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/jit/_trace.py(1184): _get_trace_graph\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/onnx/utils.py(894): _trace_and_get_graph_from_model\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/onnx/utils.py(990): _create_jit_graph\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/onnx/utils.py(1114): _model_to_graph\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/onnx/utils.py(1532): _export\n/home/prakashlab/.local/lib/python3.8/site-packages/torch/onnx/utils.py(507): export\nmake_onnx.py(25): main\nmake_onnx.py(29): <module>\n"
[03/23/2023-10:23:44] [E] [TRT] ModelImporter.cpp:776: --- End node ---
[03/23/2023-10:23:44] [E] [TRT] ModelImporter.cpp:778: ERROR: ModelImporter.cpp:180 In function parseGraph:
[6] Invalid Node - /conv4/conv4.9/conv/conv.0/Conv
/conv4/conv4.9/conv/conv.0/Conv: two inputs (data and weights) are allowed only in explicit-quantization mode.
[03/23/2023-10:23:44] [E] Failed to parse onnx file
[03/23/2023-10:23:44] [I] Finish parsing network model
[03/23/2023-10:23:44] [E] Parsing model failed
[03/23/2023-10:23:44] [E] Failed to create engine from model or file.
[03/23/2023-10:23:44] [E] Engine set up failed
&&&& FAILED TensorRT.trtexec [TensorRT v8401] # ./trtexec --onnx=/media/prakashlab/SSD/quantize_m2unet/models/flat_model_24_9_sz1024_v11.onnx --saveEngine=/media/prakashlab/SSD/quantize_m2unet/models/flat_model_24_9_sz1024_v11_from_onnx.trt --explicitBatch

AastaLLL · March 30, 2023, 5:10am

Hi,

TensorRT needs ONNX format as input. It doesn’t support the PyTorch model directly.

In your latest comment, the output error is related to the quantization process of PyTorch.
Is quantization essential for you? Do you apply quantization-aware training (QAT)?

If not, would you mind using the standard API and do you post quantization with trtexec (PTQ) instread?

Thanks.

kmarx · April 4, 2023, 4:24pm

Hi,

The NVIDIA torch2trt documentation gives instructions on how to convert from torch to TensorRT format directly; this is what I have been having trouble with.

I do not apply quantization-aware training and I haven’t been quantizing my model - you can check the code and see that everything is done in float32. I would use trtexec to do PTQ later, but for now I am still struggling with getting full-precision working.

As a sanity check, I ran torch2trt again in the l4t Docker container and got slightly different errors; I have the error attached as a txt file here:
out_torch2trt.txt (31.4 KB)

Thanks

AastaLLL · April 13, 2023, 5:05am

Hi,

Torch2TRT converts the model to TensorRT layer by layer.
So it’s possible to meet some issues if the used layer doesn’t been added/verified by the library.

In the .pth - .onnx - .trt flow on Mar 17, could you infer it with ONNXRuntime as well?
Just want to make sure if any preprocessing can be used to make the inference work.

Is the model variant of UNET?
UNET should be able to work on TensorRT.

Thanks.

kmarx · April 13, 2023, 5:15pm

Hi,

The model is a variant of UNET and I don’t think the problem is that the layers haven’t been added by the library - the errors indicate that the layers are just the wrong size rather than being incompatible with with TensorRT.

I will try using inference using the ONNXRuntime, thanks.

system · May 1, 2023, 7:07am

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.