AttributeError: module 'torch.jit' has no attribute 'get_trace_graph

[InstantWindy]

Hi ! After installing tensorwatch, I run 'demo' code, it shows "AttributeError: module 'torch.jit' has no attribute 'get_trace_graph", how to solve it ? Thanks!

[sytelus]

Could you please give more details on what exactly are you doing? What code are you executing? How do we reproduce the issue?

[InstantWindy]

Platform : Windows 7 system Version: pytorch 1.0.0, python 3.6 My code is as follows:

import torch
import torch.nn as nn
import torch.nn.init as init
import tensorwatch as tw
import torch.nn.functional as F
from torch.nn.functional import interpolate as interpolate

def channel_shuffle(x,groups):
    batchsize, num_channels, height, width = x.data.size()

    channels_per_group = num_channels // groups

    #reshape
    x = x.view(batchsize,groups,
        channels_per_group,height,width)

    x = torch.transpose(x,1,2).contiguous()

    #flatten
    x = x.view(batchsize,-1,height,width)

    return x

class Conv2dBnRelu(nn.Module):
    def __init__(self,in_ch,out_ch,kernel_size=3,stride=1,padding=0,dilation=1,bias=True):
        super(Conv2dBnRelu,self).__init__()

        self.conv = nn.Sequential(
        nn.Conv2d(in_ch,out_ch,kernel_size,stride,padding,dilation=dilation,bias=bias),
        nn.BatchNorm2d(out_ch, eps=1e-3),
        nn.ReLU(inplace=True)
    )

    def forward(self, x):
        return self.conv(x)

##after Concat -> BN, you also can use Dropout like SS_nbt_module may be make a good result!
class DownsamplerBlock (nn.Module):
    def __init__(self, in_channel, out_channel):
        super().__init__()

        self.conv = nn.Conv2d(in_channel, out_channel-in_channel, (3, 3), stride=2, padding=1, bias=True)
        self.pool = nn.MaxPool2d(2, stride=2)
        self.bn = nn.BatchNorm2d(out_channel, eps=1e-3)
        self.relu = nn.ReLU(inplace=True)

    def forward(self, input):
        output = torch.cat([self.conv(input), self.pool(input)], 1)
        output = self.bn(output)
        output = self.relu(output)
        return output 

class SS_nbt_module(nn.Module):
    def __init__(self, chann, dropprob, dilated):        
        super().__init__()

        oup_inc = chann//2

        #dw
        self.conv3x1_1_l = nn.Conv2d(oup_inc, oup_inc, (3,1), stride=1, padding=(1,0), bias=True)

        self.conv1x3_1_l = nn.Conv2d(oup_inc, oup_inc, (1,3), stride=1, padding=(0,1), bias=True)

        self.bn1_l = nn.BatchNorm2d(oup_inc, eps=1e-03)

        self.conv3x1_2_l = nn.Conv2d(oup_inc, oup_inc, (3,1), stride=1, padding=(1*dilated,0), bias=True, dilation = (dilated,1))

        self.conv1x3_2_l = nn.Conv2d(oup_inc, oup_inc, (1,3), stride=1, padding=(0,1*dilated), bias=True, dilation = (1,dilated))

        self.bn2_l = nn.BatchNorm2d(oup_inc, eps=1e-03)

        #dw
        self.conv3x1_1_r = nn.Conv2d(oup_inc, oup_inc, (3,1), stride=1, padding=(1,0), bias=True)

        self.conv1x3_1_r = nn.Conv2d(oup_inc, oup_inc, (1,3), stride=1, padding=(0,1), bias=True)

        self.bn1_r = nn.BatchNorm2d(oup_inc, eps=1e-03)

        self.conv3x1_2_r = nn.Conv2d(oup_inc, oup_inc, (3,1), stride=1, padding=(1*dilated,0), bias=True, dilation = (dilated,1))

        self.conv1x3_2_r = nn.Conv2d(oup_inc, oup_inc, (1,3), stride=1, padding=(0,1*dilated), bias=True, dilation = (1,dilated))

        self.bn2_r = nn.BatchNorm2d(oup_inc, eps=1e-03)       

        self.relu = nn.ReLU(inplace=True)
        self.dropout = nn.Dropout2d(dropprob)       

    @staticmethod
    def _concat(x,out):
        return torch.cat((x,out),1)    

    def forward(self, input):

        x1 = input[:,:(input.shape[1]//2),:,:]
        x2 = input[:,(input.shape[1]//2):,:,:]      

        output1 = self.conv3x1_1_l(x1)
        output1 = self.relu(output1)
        output1 = self.conv1x3_1_l(output1)
        output1 = self.bn1_l(output1)
        output1 = self.relu(output1)

        output1 = self.conv3x1_2_l(output1)
        output1 = self.relu(output1)
        output1 = self.conv1x3_2_l(output1)
        output1 = self.bn2_l(output1)

        output2 = self.conv1x3_1_r(x2)
        output2 = self.relu(output2)
        output2 = self.conv3x1_1_r(output2)
        output2 = self.bn1_r(output2)
        output2 = self.relu(output2)

        output2 = self.conv1x3_2_r(output2)
        output2 = self.relu(output2)
        output2 = self.conv3x1_2_r(output2)
        output2 = self.bn2_r(output2)

        if (self.dropout.p != 0):
            output1 = self.dropout(output1)
            output2 = self.dropout(output2)

        out = self._concat(output1,output2)
        out = F.relu(input+out,inplace=True)
        return channel_shuffle(out,2)     

class Encoder(nn.Module):
    def __init__(self, num_classes):
        super().__init__()
        self.initial_block = DownsamplerBlock(3,32)

        self.layers = nn.ModuleList()

        for x in range(0, 3):   
           self.layers.append(SS_nbt_module(32, 0.03, 1)) 

        self.layers.append(DownsamplerBlock(32,64))

        for x in range(0, 2):   
           self.layers.append(SS_nbt_module(64, 0.03, 1)) 

        self.layers.append(DownsamplerBlock(64,128))

        for x in range(0, 1):    
            self.layers.append(SS_nbt_module(128, 0.3, 1))
            self.layers.append(SS_nbt_module(128, 0.3, 2))
            self.layers.append(SS_nbt_module(128, 0.3, 5))
            self.layers.append(SS_nbt_module(128, 0.3, 9))

        for x in range(0, 1):    
            self.layers.append(SS_nbt_module(128, 0.3, 2))
            self.layers.append(SS_nbt_module(128, 0.3, 5))
            self.layers.append(SS_nbt_module(128, 0.3, 9))
            self.layers.append(SS_nbt_module(128, 0.3, 17))

        #Only in encoder mode:
        self.output_conv = nn.Conv2d(128, num_classes, 1, stride=1, padding=0, bias=True)

    def forward(self, input, predict=False):

        output = self.initial_block(input)

        for layer in self.layers:
            output = layer(output)

        if predict:
            output = self.output_conv(output)

        return output

class Interpolate(nn.Module):
    def __init__(self,size,mode):
        super(Interpolate,self).__init__()

        self.interp = nn.functional.interpolate
        self.size = size
        self.mode = mode
    def forward(self,x):
        x = self.interp(x,size=self.size,mode=self.mode,align_corners=True)
        return x

class APN_Module(nn.Module):
    def __init__(self, in_ch, out_ch):
        super(APN_Module, self).__init__()      
    # global pooling branch
        self.branch1 = nn.Sequential(
                nn.AdaptiveAvgPool2d(1),
                Conv2dBnRelu(in_ch, out_ch, kernel_size=1, stride=1, padding=0)
    )

    # midddle branch
        self.mid = nn.Sequential(
        Conv2dBnRelu(in_ch, out_ch, kernel_size=1, stride=1, padding=0)
    )

        self.down1 = Conv2dBnRelu(in_ch, 1, kernel_size=7, stride=2, padding=3)

        self.down2 = Conv2dBnRelu(1, 1, kernel_size=5, stride=2, padding=2)

        self.down3 = nn.Sequential(
        Conv2dBnRelu(1, 1, kernel_size=3, stride=2, padding=1),
        Conv2dBnRelu(1, 1, kernel_size=3, stride=1, padding=1)
    )

        self.conv2 = Conv2dBnRelu(1, 1, kernel_size=5, stride=1, padding=2)
        self.conv1 = Conv2dBnRelu(1, 1, kernel_size=7, stride=1, padding=3)

    def forward(self, x):

        h = x.size()[2]
        w = x.size()[3]

        b1 = self.branch1(x)
        #b1 = Interpolate(size=(h, w), mode="bilinear")(b1)
        b1= interpolate(b1, size=(h, w), mode="bilinear", align_corners=True)

        mid = self.mid(x)

        x1 = self.down1(x)
        x2 = self.down2(x1)
        x3 = self.down3(x2)
        #x3 = Interpolate(size=(h // 4, w // 4), mode="bilinear")(x3)
        x3= interpolate(x3, size=(h // 4, w // 4), mode="bilinear", align_corners=True) 
        x2 = self.conv2(x2)
        x = x2 + x3
        #x = Interpolate(size=(h // 2, w // 2), mode="bilinear")(x)
        x= interpolate(x, size=(h // 2, w // 2), mode="bilinear", align_corners=True)

        x1 = self.conv1(x1)
        x = x + x1
        #x = Interpolate(size=(h, w), mode="bilinear")(x)
        x= interpolate(x, size=(h, w), mode="bilinear", align_corners=True)

        x = torch.mul(x, mid)

        x = x + b1

        return x

class Decoder (nn.Module):
    def __init__(self, num_classes):
        super().__init__()

        self.apn = APN_Module(in_ch=128,out_ch=20)
        #self.upsample = Interpolate(size=(512, 1024), mode="bilinear")
        #self.output_conv = nn.ConvTranspose2d(16, num_classes, kernel_size=4, stride=2, padding=1, output_padding=0, bias=True)
        #self.output_conv = nn.ConvTranspose2d(16, num_classes, kernel_size=3, stride=2, padding=1, output_padding=1, bias=True)
        #self.output_conv = nn.ConvTranspose2d(16, num_classes, kernel_size=2, stride=2, padding=0, output_padding=0, bias=True)

    def forward(self, input):

        output = self.apn(input)
        out = interpolate(output, size=(512, 1024), mode="bilinear", align_corners=True)
        #out = self.upsample(output)
        return out

# LEDNet
class Net(nn.Module):
    def __init__(self, num_classes, encoder=None):  
        super().__init__()

        if (encoder == None):
            self.encoder = Encoder(num_classes)
        else:
            self.encoder = encoder
        self.decoder = Decoder(num_classes)

    def forward(self, input, only_encode=False):
        if only_encode:
            return self.encoder.forward(input, predict=True)
        else:
            output = self.encoder(input)    
            return self.decoder.forward(output)

net = Net(20)
tw.draw_model(net,[2,3,512,1024])

[romesco]

I also run into this issue when running the example: https://github.com/microsoft/tensorwatch/blob/master/notebooks/network_arch.ipynb

I'm on: ubuntu 16.0.4.6, python3.8 and here's my pip freeze output:

attrs==19.3.0
backcall==0.1.0
bleach==3.1.1
cycler==0.10.0
decorator==4.4.1
defusedxml==0.6.0
dill==0.3.1.1
entrypoints==0.3
fvcore==0.1.dev200218
graphviz==0.13.2
imageio==2.8.0
ipykernel==5.1.4
ipython==7.12.0
ipython-genutils==0.2.0
ipywidgets==7.5.1
jedi==0.16.0
Jinja2==2.11.1
joblib==0.14.1
jsonschema==3.2.0
jupyter-client==5.3.4
jupyter-core==4.6.3
kiwisolver==1.1.0
MarkupSafe==1.1.1
matplotlib==3.1.3
mistune==0.8.4
nbconvert==5.6.1
nbformat==5.0.4
networkx==2.4
notebook==6.0.3
numpy==1.18.1
pandas==1.0.1
pandocfilters==1.4.2
parso==0.6.1
pexpect==4.8.0
pickleshare==0.7.5
Pillow==7.0.0
plotly==4.5.1
portalocker==1.5.2
prometheus-client==0.7.1
prompt-toolkit==3.0.3
ptyprocess==0.6.0
pydot==1.4.1
Pygments==2.5.2
pyparsing==2.4.6
pyrsistent==0.15.7
python-dateutil==2.8.1
pytorch3d==0.1
pytz==2019.3
PyWavelets==1.1.1
PyYAML==5.3
pyzmq==18.1.1
receptivefield==0.5.0
retrying==1.3.3
scikit-image==0.16.2
scikit-learn==0.22.1
scipy==1.4.1
Send2Trash==1.5.0
six==1.14.0
sklearn==0.0
tensorwatch==0.9.0
termcolor==1.1.0
terminado==0.8.3
testpath==0.4.4
torch==1.4.0
torchstat==0.0.7
torchvision==0.5.0
tornado==6.0.3
tqdm==4.43.0
traitlets==4.3.3
wcwidth==0.1.8
webencodings==0.5.1
widgetsnbextension==3.5.1
yacs==0.1.6

[eecheonwu]

i'm also having the same issue with the file mentioned by romesc-CMU: tw.draw_model(alexnet_model, [1, 3, 224, 224])

---

AttributeError Traceback (most recent call last)

in ----> 1 tw.draw_model(alexnet_model, [1, 3, 224, 224]) ~/.local/lib/python3.7/site-packages/tensorwatch-0.9.0-py3.7.egg/tensorwatch/__init__.py in draw_model(model, input_shape, orientation, png_filename) 33 def draw_model(model, input_shape=None, orientation='TB', png_filename=None): #orientation = 'LR' for landscpe 34 from .model_graph.hiddenlayer import pytorch_draw_model ---> 35 g = pytorch_draw_model.draw_graph(model, input_shape) 36 return g 37 ~/.local/lib/python3.7/site-packages/tensorwatch-0.9.0-py3.7.egg/tensorwatch/model_graph/hiddenlayer/pytorch_draw_model.py in draw_graph(model, args) 33 args = torch.ones(args) 34 ---> 35 dot = draw_img_classifier(model, args) 36 return DotWrapper(dot) 37 ~/.local/lib/python3.7/site-packages/tensorwatch-0.9.0-py3.7.egg/tensorwatch/model_graph/hiddenlayer/pytorch_draw_model.py in draw_img_classifier(model, dataset, display_param_nodes, rankdir, styles, input_shape) 61 try: 62 non_para_model = distiller.make_non_parallel_copy(model) ---> 63 g = SummaryGraph(non_para_model, dummy_input) 64 65 return sgraph2dot(g, display_param_nodes, rankdir, styles) ~/.local/lib/python3.7/site-packages/tensorwatch-0.9.0-py3.7.egg/tensorwatch/model_graph/hiddenlayer/summary_graph.py in __init__(self, model, dummy_input, apply_scope_name_workarounds) 88 dummy_input = distiller.convert_tensors_recursively_to(dummy_input, device=device) 89 self.dummy_input = dummy_input ---> 90 trace, _ = jit.get_trace_graph(model_clone, dummy_input, _force_outplace=True) 91 92 # As of PyTorch 1.1.0, ONNX trace optimization has two issues that result in incorrect scope names AttributeError: module 'torch.jit' has no attribute 'get_trace_graph'

[stu10064]

I'm also running the same issue. I think after pytorch 1.1, get_trace_graph function is removed.

[sytelus]

The internals of PyTorch is certainly changed after 1.1. I have fixed the error in the latest commit but now the graph has many extraneous nodes so this still needs to be investigated.

[LexTran]

Hey, I have met the same problem. And I looked at the new version source code of torch.jit, compared with the old version, it changes the function name from get_trace_graph to _get_trace_graph. So I changed the function in my code, although there's other problem, but it didn't show AttributeError: module 'torch.jit' has no attribute 'get_trace_graph', try it, it may work. Hope this can help you.

[H19012]

Hey, I have met the same problem. And I looked at the new version source code of torch.jit, compared with the old version, it changes the function name from get_trace_graph to _get_trace_graph. So I changed the function in my code, although there's other problem, but it didn't show AttributeError: module 'torch.jit' has no attribute 'get_trace_graph', try it, it may work. Hope this can help you.

Now showing AttributeError: 'torch._C.Graph' object has no attribute 'set_graph' for me.

[JobAtom]

Hey, I have met the same problem. And I looked at the new version source code of torch.jit, compared with the old version, it changes the function name from get_trace_graph to _get_trace_graph. So I changed the function in my code, although there's other problem, but it didn't show AttributeError: module 'torch.jit' has no attribute 'get_trace_graph', try it, it may work. Hope this can help you.

Now showing AttributeError: 'torch._C.Graph' object has no attribute 'set_graph' for me.

Me too. Do you solve this problem?

[firestonelib]

Hey, I have met the same problem. And I looked at the new version source code of torch.jit, compared with the old version, it changes the function name from get_trace_graph to _get_trace_graph. So I changed the function in my code, although there's other problem, but it didn't show AttributeError: module 'torch.jit' has no attribute 'get_trace_graph', try it, it may work. Hope this can help you.

Now showing AttributeError: 'torch._C.Graph' object has no attribute 'set_graph' for me.

My solution is using from torch.utils.tensorboard import SummaryWriterrather than from tensorboardX import SummaryWriter, all you need to do is to modify the script 'python3.7/site-packages/torch/jit/init.py' at line 1884:

            try:
                tmp_module._modules[name] = make_module(submodule, TracedModule, _compilation_unit=None)
            except:
                continue 

[isekeylabSunQixiang]

For HRNet you can change this code writer_dict['writer'].add_graph(model, (dump_input, )) to # writer_dict['writer'].add_graph(model, (dump_input, ))