search

epfml / sent2vec

General purpose unsupervised sentence representations
Other
1.19k stars 256 forks source link

Training process fault #45

Closed darentsia closed 4 years ago

darentsia commented 6 years ago

../sent2vec/fasttext sent2vec -input data/preprocessed_docs/result_0.txt -output my_model_s2v -minCount 10 -dim 700 -epoch 7 -lr 0.3 -wordNgrams 2 -loss ns -neg 10 -thread 48 -t 0.00005 -dropoutK 4 -bucket 10000
Read 227M words Number of words: 374566 Number of labels: 0 Progress: 19.0% words/sec/thread: 59 lr: 0.242889 loss: 20.702034 eta: 125h56m Segmentation fault

Trained with these parameters, but the training process ended with the message:

"Segmentation fault"

Can you, please, explain, why?

Machine parameters: 48 cores, 126Gb RAM. 1.5Tb of free disk space.

Size of the input file: 1.5Gb

guptaprkhr commented 6 years ago

Hi, I think you should try lowering the learning rate. Sometimes training with a high learning rate might end up blowing up the gradients.