convert_torch_to_pytorch
convert_torch_to_pytorch copied to clipboard
clcarwin
convert torch 7 model and get an error :AttributeError: type object 'torch.cuda.FloatStorage' has no attribute 'from_buffer'
wangjian@lhtserver-2:~/wj$ python convert_torch.py -m vgg.t7
Traceback (most recent call last):
File "convert_torch.py", line 286, in
Where did you download the vgg.t7? You can try check the size of the vgg.t7 and convert torch.cuda.FloatStorage to torch.FloatStorage before using it.
vgg model from https://github.com/jcjohnson/cnn-benchmarks has been tested, you can use this version.
@clcarwin I got the same error. I try to use this library to convert a standalone training model (from my experiments; not vgg). So is that mean I cannot use this library to convert torch.cuda.FloatStorage?
@johnny5550822 convert GPU to CPU model first.
m = torch.load('xxxx.t7')
m = m:float()
torch.save('xxxx.cpu.t7',m)
Thanks ;) worked for resnet152 from https://github.com/facebook/fb.resnet.torch
@johnny5550822 do you get the problem " 'torch.FloatTensor' object has no attribute 'output' ",how should I do?
@heyalqh did not get this problem. Can you share the screenshot of the code and error?
Does this mean that I cannot use the model on the GPU after conversion?
When I do module.cuda() and attempt to evaluate module.forward(data), I get the error:
RuntimeError: expected CPU tensor (got CUDA tensor)
When I put my data on the GPU data = Variable(torch.from_numpy(batch)).cuda()
I end up with this error instead:
Traceback (most recent call last):
File "thing.py", line 51, in <module>
module.forward(data.float())
File "/usr/local/lib/python2.7/dist-packages/torch/nn/modules/container.py", line 64, in forward
input = module(input)
File "/usr/local/lib/python2.7/dist-packages/torch/nn/modules/module.py", line 206, in __call__
result = self.forward(*input, **kwargs)
File "/opt/home/d2curro/repos/thinslicing_pytorch/cpu_net.py", line 21, in forward
return self.lambda_func(self.forward_prepare(input))
File "/opt/home/d2curro/repos/thinslicing_pytorch/cpu_net.py", line 35, in <lambda>
Lambda(lambda x,lrn=torch.legacy.nn.SpatialCrossMapLRN(*(5, 0.0005, 0.75, 2)): Variable(lrn.forward(x.data))),
File "/usr/local/lib/python2.7/dist-packages/torch/legacy/nn/Module.py", line 33, in forward
return self.updateOutput(input)
File "/usr/local/lib/python2.7/dist-packages/torch/legacy/nn/SpatialCrossMapLRN.py", line 25, in updateOutput
self._backend.SpatialCrossMapLRN_updateOutput(
File "/usr/local/lib/python2.7/dist-packages/torch/_thnn/utils.py", line 22, in __getattr__
raise NotImplementedError
NotImplementedError
when I try to convert the following torch net to pytorch,
require 'torch' require 'nn' require 'nnx' require 'optim' require 'rnn'
function buildModel_MeanPool_RNN(nFltrs1,nFltrs2,nFltrs3,nPersonsTrain)
local nFilters = {nFltrs1,nFltrs2,nFltrs3}
local filtsize = {5,5,5} local poolsize = {2,2,2} local stepSize = {2,2,2}
-- remember this adds padding to ALL SIDES of the image local padDim = 4
local cnn = nn.Sequential()
local ninputChannels = 5 cnn:add(nn.SpatialZeroPadding(padDim, padDim, padDim, padDim)) cnn:add(nn.SpatialConvolutionMM(ninputChannels, nFilters[1], filtsize[1], filtsize[1], 1, 1)) cnn:add(nn.Tanh()) cnn:add(nn.SpatialMaxPooling(poolsize[1],poolsize[1],stepSize[1],stepSize[1]))
ninputChannels = nFilters[1] cnn:add(nn.SpatialZeroPadding(padDim, padDim, padDim, padDim)) cnn:add(nn.SpatialConvolutionMM(ninputChannels, nFilters[2], filtsize[2], filtsize[2], 1, 1)) cnn:add(nn.Tanh()) cnn:add(nn.SpatialMaxPooling(poolsize[2],poolsize[2],stepSize[2],stepSize[2]))
ninputChannels = nFilters[2] cnn:add(nn.SpatialZeroPadding(padDim, padDim, padDim, padDim)) cnn:add(nn.SpatialConvolutionMM(ninputChannels, nFilters[3], filtsize[3], filtsize[3], 1, 1)) cnn:add(nn.Tanh()) cnn:add(nn.SpatialMaxPooling(poolsize[3],poolsize[3],stepSize[3],stepSize[3]))
local nFullyConnected = nFilters[3]108
cnn:add(nn.Reshape(1,nFullyConnected))
cnn:add(nn.Dropout(0.6))
cnn:add(nn.Linear(nFullyConnected,128))
-- cnn:cuda()
local h2h = nn.Sequential() h2h:add(nn.Tanh()) h2h:add(nn.Dropout(0.6)) h2h:add(nn.Linear(128,128)) -- h2h:cuda()
local r1 = nn.Recurrent( 128, cnn, h2h, nn.Identity(), 16)
local rnn1 = nn.Sequencer( nn.Sequential() :add(r1) )
Combined_CNN_RNN_1 = nn.Sequential() Combined_CNN_RNN_1:add(rnn1) Combined_CNN_RNN_1:add(nn.JoinTable(1)) Combined_CNN_RNN_1:add(nn.Mean(1))
local r2 = nn.Recurrent( 128, cnn:clone('weight','bias','gradWeight','gradBias'), h2h:clone('weight','bias','gradWeight','gradBias'), nn.Identity(), 16)
local rnn2 = nn.Sequencer( nn.Sequential() :add(r2) )
Combined_CNN_RNN_2 = nn.Sequential() Combined_CNN_RNN_2:add(rnn2) Combined_CNN_RNN_2:add(nn.JoinTable(1)) Combined_CNN_RNN_2:add(nn.Mean(1))
-- Combined_CNN_RNN_2 = Combined_CNN_RNN_1:clone('weight','bias','gradWeight','gradBias')
local mlp2 = nn.ParallelTable() mlp2:add(Combined_CNN_RNN_1) mlp2:add(Combined_CNN_RNN_2) -- mlp2:cuda()
local mlp3 = nn.ConcatTable() mlp3:add(nn.Identity()) mlp3:add(nn.Identity()) mlp3:add(nn.Identity()) -- mlp3:cuda()
local mlp4 = nn.ParallelTable() mlp4:add(nn.Identity()) mlp4:add(nn.SelectTable(1)) mlp4:add(nn.SelectTable(2)) -- mlp4:cuda()
-- used to predict the identity of each person local classifierLayer = nn.Linear(128,nPersonsTrain)
-- identification local mlp6 = nn.Sequential() mlp6:add(classifierLayer) mlp6:add(nn.LogSoftMax()) -- mlp6:cuda()
local mlp7 = nn.Sequential() mlp7:add(classifierLayer:clone('weight','bias','gradWeight','gradBias')) mlp7:add(nn.LogSoftMax()) -- mlp7:cuda()
local mlp5 = nn.ParallelTable() mlp5:add(nn.PairwiseDistance(2)) mlp5:add(mlp6) mlp5:add(mlp7) -- mlp5:cuda()
local fullModel = nn.Sequential() fullModel:add(mlp2) fullModel:add(mlp3) fullModel:add(mlp4) fullModel:add(mlp5) -- fullModel:cuda()
local crit = nn.SuperCriterion() crit:add(nn.HingeEmbeddingCriterion(2),1) crit:add(nn.ClassNLLCriterion(),1) crit:add(nn.ClassNLLCriterion(),1)
return fullModel, crit, Combined_CNN_RNN_1, cnn end
fullModel, crit, Combined_CNN_RNN_1, cnn = buildModel_MeanPool_RNN(16,32,32,150) torch.save('Combined_CNN_RNN.t7',Combined_CNN_RNN_1)
But it is not working: Not Implement nn.Sequencer Not Implement JoinTable Not Implement Mean
@clcarwin , when i try to convert the model to use float tensors (the original model was trained on GPUs and CUDA tensors is not usable in the code), I get this error:
[string "m = m:float()"]:1: attempt to call method 'float' (a nil value)
stack traceback:
[string "m = m:float()"]:1: in main chunk
[C]: in function 'xpcall'
/home/dravi/torch/install/share/lua/5.1/trepl/init.lua:679: in function 'repl'
...ravi/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:204: in main chunk
[C]: at 0x00405d50
I'm doing this via the command line 'th', is there any package that is missing and needs to be 'required' for the float tensor conversion?
@stalagmite7 @huntkao I am having the same issue! Did either of you end up solving this issue?
Thanks a lot!
No, @huntkao and @Spandan-Madan . I finally ended up not going down this route because it was so buggy
@clcarwin Hi, I have SyntaxError on the line "m = m:float()". How to solve it? Thanks a lot!
In my specific case, this is what happened:-
lua uses tables for all data arrays pretty much. So the only way to find out if the object is a model itself or a table containing model and other things is to look inside the table and see what it contains. In my case, the table object was a list of tables and one of the tables inside it was the model. So, I indexed inside to the right element. And then converted that model to CPU using model = model:float() and it worked perfectly.
I know it sounds silly, but give it a shot. You never know you might have the same problem as me!
On Sat, Aug 18, 2018 at 6:10 PM, dmortem [email protected] wrote:
@clcarwin https://github.com/clcarwin Hi, I have SyntaxError on the line "m = m:float()". How to solve it? Thanks a lot!
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/clcarwin/convert_torch_to_pytorch/issues/8#issuecomment-414096009, or mute the thread https://github.com/notifications/unsubscribe-auth/AU8Xo45ExiBYx1z0Y8ItlYXr2WTROPgGks5uSLsDgaJpZM4NjX48 .
@dmortem Hi, I have the same error with you, do you overcome it and how?
@dmortem Thanks, do you change torch to pytorch successfully? I met error as @stalagmite7. And I try to download xnornet.py but it still show error "string m=m:float() attempt to call method 'float' (a nil value) ".
@LaiPiXiong After I configuring torch environment, I put three sentences in a lua file, and execute it. The model has been converted successfully. And whether you can use this model in Pytorch depends on the original torch model since some torch structures cannot be used in Pytorch, so maybe we still cannot use the converted model.
@dmortem I can't convert it right. Does "convert_torch.py" also based on torch?
@LaiPiXiong I'm sorry, I didn't convert the model successfully, and my error is "torch.utils.serialization.read_lua_file.T7ReaderException: unknown type id 436207616. The file may be corrupted.". This error is because my model has some different layers which are absent in torch.legacy. And what I do is to run lua file to convert the model of the GPU version into the CPU version, and then run python file to convert it into a pytorch model.
@dmortem I can't convert model in GPU to CPU version. Do you install luarocks before torch?
@dmortem Hi, I convert torch model to pytorch successfully. I use 'import 'cudnn'' before "m = torch.load(model-name)". I learn this from https://github.com/rwightman/pytorch-planet-amazon/issues/1
@johnny5550822 You should download the model from https://github.com/jcjohnson/cnn-benchmarks This does not contain the cudnn contents and can be successful.
I am trying to convert this model driveLink from this repo repoLink but it gives the error
return python_class.from_buffer(reader.f.read(size), 'native') AttributeError: type object 'torch.cuda.FloatStorage' has no attribute 'from_buffer'
i tried to convert it to cpu but the model wont load using torch with error magic_number = pickle_module.load(f) _pickle.UnpicklingError: invalid load key, '\x03'.
any help will be appreciated
