search

macournoyer / neuralconvo

Neural conversational model in Torch
777 stars 346 forks source link

Not enough memory? #5

Closed llSourcell closed 8 years ago

llSourcell commented 8 years ago

I ran this on the CPU for a few hours and when saving the model after training I got an error. Any fix ideas? I'm running this on a macbook with 16 GB of RAM

1482-sraval:neuralconvo-master sraval$ th train.lua --dataset 5000 --hiddenSize 1000
-- Loading dataset  
Loading vocabulary from data/vocab.t7 ...   

Dataset stats:  
  Vocabulary size: 7061 
         Examples: 8151 

-- Epoch 1 / 50 

 [=========================================== 8151/8151 h43m | Step: 690ms      

Finished in 1h43m 1.3066972012208 examples/sec. 

Epoch stats:    
           LR= 0.05 
  Errors: min= 1.1344377487802  
          max= 204.85985171145  
       median= 25.911087689104  
         mean= 33.523359973026  
          std= 27.159547015804  

(Saving model ...)  
/Users/sraval/torch/install/bin/luajit: not enough memory
macournoyer commented 8 years ago

That is weird... I think 16 GB is enough. Are you using lots of RAM for other apps at the same time?

Also, if your MBP is recent, you should be able to setup CUDA which will reduce training time to ~10min per epoch and might help w/ mem issue.

kalloc commented 8 years ago

I'm thinking this error from CUDA ( MBPr NVIDIA doesn't have enough memory)

macournoyer commented 8 years ago

Are you using your computer for other things while this is running?