search

harvardnlp / seq2seq-attn

Sequence-to-sequence model with LSTM encoder/decoders and attention
http://nlp.seas.harvard.edu/code
MIT License
1.26k stars 278 forks source link

Train from issue #20

Closed penguinshin closed 8 years ago

penguinshin commented 8 years ago

Hi, sorry for not replying! I am still having the same issue where loading a model gives the empty tensor error. have you had this issue still?

yoonkim commented 8 years ago

if I recall correctly I think the previous train_from error was not an empty tensor error. Can you post your command line as well as the error it gives?

penguinshin commented 8 years ago

Thanks for responding yoon! I just looked at the previous post I made and it was an empty tensor error.

Here is the new one. Keep in mind that my run_model2.py encodes all the command line things.

Would i be able to send you this file over email?

Austins-MacBook-Pro:src2anno austinshin$ python run_model2.py loading data... done!
Source vocab size: 8814, Target vocab size: 15669
Source max sent len: 50, Target max sent len: 52
loading django-model_final1.t7...
Number of parameters: 14535669
/Users/austinshin/torch/install/bin/luajit: bad argument #1 to '?' (empty tensor at /Users/austinshin/torch/pkg/torch/generic/Tensor.c:888) stack traceback: [C]: at 0x0f9afad0 [C]: in function '__index' /Users/austinshin/torch/install/share/lua/5.1/nn/MM.lua:51: in function 'updateGradInput' ...stinshin/torch/install/share/lua/5.1/nngraph/gmodule.lua:386: in function 'neteval' ...stinshin/torch/install/share/lua/5.1/nngraph/gmodule.lua:420: in function 'updateGradInput' ...stinshin/torch/install/share/lua/5.1/nngraph/gmodule.lua:386: in function 'neteval' ...stinshin/torch/install/share/lua/5.1/nngraph/gmodule.lua:420: in function 'updateGradInput' /Users/austinshin/torch/install/share/lua/5.1/nn/Module.lua:31: in function 'backward' train.lua:370: in function 'train_batch' train.lua:479: in function 'train' train.lua:646: in function 'main' train.lua:649: in main chunk [C]: in function 'dofile' ...shin/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk [C]: at 0x010f647bc0

yoonkim commented 8 years ago

ah ok. yes, please send the file over email!

penguinshin commented 8 years ago

thanks!

On May 1, 2016, at 10:35 PM, Yoon Kim notifications@github.com wrote:

ah ok. yes, please send the file over email!

— You are receiving this because you authored the thread. Reply to this email directly or view it on GitHub https://github.com/harvardnlp/seq2seq-attn/issues/20#issuecomment-216094706

penguinshin commented 8 years ago

let me know if you get the file. im not sure if it sent

yoonkim commented 8 years ago

hmm i didn't get it...

penguinshin commented 8 years ago

here it is as a text file, just change the extension to py run_model2.txt

yoonkim commented 8 years ago

hmm can you add the following two lines and print out the log here?

line 616: print(opt.num_layers, model_opt.num_layers) line 369: print(#drnn_state_dec)

penguinshin commented 8 years ago

Austins-MacBook-Pro:src2anno austinshin$ python run_model2.py loading data... done!
Source vocab size: 8814, Target vocab size: 15669
Source max sent len: 50, Target max sent len: 52
loading django-model_final1.t7...
2 2
Number of parameters: 14535669
5
/Users/austinshin/torch/install/bin/luajit: bad argument #1 to '?' (empty tensor at /Users/austinshin/torch/pkg/torch/generic/Tensor.c:888) stack traceback:

On May 1, 2016, at 10:50 PM, Yoon Kim notifications@github.com wrote:

print(opt.num_layers, model_opt.num_layers)

yoonkim commented 8 years ago

huh, weird. that looks right. let's debug this in person tomorrow? i should be in my office most of the afternoon

penguinshin commented 8 years ago

sounds good. wheres your office?

On May 1, 2016, at 11:17 PM, Yoon Kim notifications@github.com wrote:

print(opt.num_layers, model_opt.num_layers)

yoonkim commented 8 years ago

MD208

penguinshin commented 8 years ago

I printed out decoder_input (line 370) does this look right?

Austins-MacBook-Pro:src2anno austinshin$ python run_model2.py loading data... done!
Source vocab size: 8814, Target vocab size: 15669
Source max sent len: 50, Target max sent len: 52
loading django-model_final1.t7...
2 2
Number of parameters: 14535669
5
{ 1 : LongTensor - size: 64 2 : DoubleTensor - size: 64x12x100 3 : DoubleTensor - size: 64x100 4 : DoubleTensor - size: 64x100 5 : DoubleTensor - size: 64x100 6 : DoubleTensor - size: 64x100 7 : DoubleTensor - size: 64x100 } /Users/austinshin/torch/install/bin/luajit: bad argument #1 to '?' (empty tensor at /Users/austinshin/torch/pkg/torch/generic/Tensor.c:888) stack traceback:

It’s odd because when I run a new training (when i don’t have the train_from) I get this (along with no error):

{ 1 : LongTensor - size: 64 2 : DoubleTensor - size: 64x6x100 3 : DoubleTensor - size: 64x100 4 : DoubleTensor - size: 64x100 5 : DoubleTensor - size: 64x100 6 : DoubleTensor - size: 64x100 7 : DoubleTensor - size: 64x100 }

penguinshin commented 8 years ago

So for some reason, when I do train_from, the value of source_l is 12 instead of 6 for normal training (without train_from). i think this might be the issue

On May 1, 2016, at 11:18 PM, Yoon Kim notifications@github.com wrote:

MD208

— You are receiving this because you authored the thread. Reply to this email directly or view it on GitHub https://github.com/harvardnlp/seq2seq-attn/issues/20#issuecomment-216098419

penguinshin commented 8 years ago

Hi yoon I just stopped by your office- will Be in md for the next hour or so let me know if your here!

Sent from my iPhone

On May 1, 2016, at 11:18 PM, Yoon Kim notifications@github.com wrote:

MD208

— You are receiving this because you authored the thread. Reply to this email directly or view it on GitHub

yoonkim commented 8 years ago

hey sorry im in my office now!

Sarah20187 commented 6 years ago

How do you fix this problem? I got a same one?