mtanana / torchneuralconvo Goto Github PK

I've been working on some revamps for this project that make it much easier to restart models that have stopped from unexpected events (i.e. computer crashes, or you were preempted on some shared resources) . I think this changes might make it tricky to load any saved models you have right now.

So I'm going to hold off on the commit for a few weeks to make sure people are aware.

But be warned!

tried running your code without cuda
stopped at 7 epoch cycles
loaded a new dataset received this error please help
I am new to lua programming

could you please help me understand the error

is it because of large input and output string I entered

-- Loading dataset
vocab.t7 or data.t7 detected as stale. Re-running preprocessing...
one-time setup: preprocessing input text file data/opensubssmall/input.txt...
loading text file...
creating vocabulary mapping...
1027
Count: 385
Length: 1027
putting data into tensor...
/home/guru99u2/torch/install/bin/luajit: ./util/WordSplitLMMinibatchLoader.lua:529: bad argument #3 to 'narrow' (out of range at /tmp/luarocks_torch-scm-1-405/torch7/lib/TH/generic/THTensor.c:351)
stack traceback:
[C]: in function 'narrow'
./util/WordSplitLMMinibatchLoader.lua:529: in function 'text_to_tensor'
./util/WordSplitLMMinibatchLoader.lua:90: in function 'create'
trainoptim.lua:67: in main chunk
[C]: in function 'dofile'
...99u2/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
[C]: at 0x00406670

is there any way to get this working with a AMD GPU using opencl or something like it?

I have multiple questions for you.

1. You are reversing the input sequence during training but i don't see you doing this with the input sequence when using eval.lua . Am I missing something?

2. You mention a beam search implementation but there seems to be no beam.lua. Could you please add it ?

3. Could you give a little bit more information about the dataset you used ? What do you mean by the first 9 milion ? And could you please share some information about the errors you achieved in the training?

Thank you!