Train Error on GPU ClassNLLCriterion.lua:34: bad argument #1 (field weights does not exist) about char-rnn HOT 10 CLOSED

I also experienced this error. Unfortunately, I have been poking around quite a bit in an effort to solve an unrelated issue.

If I recall correctly, I tried luarocks install nn, which may have fixed it, then luarocks install torch, which fixed not being able to find the cunn and cutorch libraries. I'm fairly certain the latter command is going to come back to haunt me: installing torch through luarocks over an existing torch install doesn't seem very clean

from char-rnn.

Everything goes well now. first I update the torch by update.sh, second I link the cuda to cuda-7.5, third I run the torch's ./install.sh again. then everything goes well. But I still can't figure the problem. I just resolve this problem by accident. Hope this can help somebody.
I will close this issue.

from char-rnn.

fwiw I've seen this recently too

from char-rnn.

see also torch/nn@6bcdc6f ... I fixed an installation by brutishly copying in latest version of that class over the top of luajit. #notproud

from char-rnn.

I'm not sure what's the cause for this issue or when it started. Is it cuda version dependent?

from char-rnn.

I can confirm that I did update CUDA before this happened, but I can't confirm that the update was the cause, as I twiddled with too many other things in that time

from char-rnn.

the nn and cunn versions looks out of date. that's what this error message is about.

from char-rnn.

Just writing down my fix for the same error:

user@xyz:~/code/char-rnn$ th train.lua -data_dir data/tinyshakespeare
using CUDA on GPU 0...
vocab.t7 and data.t7 do not exist. Running preprocessing...
one-time setup: preprocessing input text file data/tinyshakespeare/input.txt...
loading text file...
creating vocabulary mapping...
putting data into tensor...
saving data/tinyshakespeare/vocab.t7
saving data/tinyshakespeare/data.t7
loading data files...
cutting off end of data so that the batches/sequences divide evenly
reshaping tensor...
data load done. Number of data batches in train: 423, val: 23, test: 0
vocab size: 65
creating an lstm with 2 layers
number of parameters in the model: 240321
cloning rnn
cloning criterion
/Users/user/code/torch/install/bin/luajit: ...ode/torch/install/share/lua/5.1/nn/ClassNLLCriterion.lua:50: bad argument #1 to 'ClassNLLCriterion_updateOutput' (torch.CudaTensor expected, got nn.ClassNLLCriterion)
stack traceback:
    [C]: in function 'ClassNLLCriterion_updateOutput'
    ...ode/torch/install/share/lua/5.1/nn/ClassNLLCriterion.lua:50: in function 'forward'
    train.lua:254: in function 'opfunc'
    .../user/code/torch/install/share/lua/5.1/optim/rmsprop.lua:32: in function 'rmsprop'
    train.lua:293: in main chunk
    [C]: in function 'dofile'
    ...code/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:131: in main chunk
    [C]: at 0x010e40f2e0

I am using Cuda 7.0 on NVIDIA GeForce GT 650M on my Macbook Pro

Updating torch alone did not help.

As suggested by @raidancampbell I've run luarocks install nn and it resolved the problem.

from char-rnn.

+1 luarocks install nn worked for me too. PR - #92

from char-rnn.

the changes the current nn need to be merged into getParamsByDevice otherwise data parallel wont work

from char-rnn.