Taking too long for Trie


(SGang) #1

I have a 18 GB binary file. The time being taken for generating a trie file is well over 10 hours now and its yet to be completed. There’s nothing being shown on the screen either.

I’ve about 150GB of hard disc space and a 16GB DDR4 RAM (I dont think there’s any use of GPU here…!)

How long should it take and what should I do to reduce the time taken> ANd is this normal?

Any help would be greatly appreciated.


(Dan) #2

Not familiar with generating a trie, but an 18GB binary file may take quite some time to process using a hard disk, hence why Mozilla recommended using a SSD. iotop and htop can likely give you an insight as to whether your I/O limited (hard drive is too slow), or the CPU or RAM are your bottleneck.


(SGang) #3

I understand Dan. I stopped the cycle. But there’s some more issues now… :). Adding a separate question.


(Lissyx) #4

Is this lm.binary ? That feels way too big, and given your system, it’s not unexpected to be that slow.

How did you built this one ?

How long it would take is highly dependang on your system itself, it’s hard to give a ballpark without more informations.


(SGang) #5

It was done. Not sure if it was right or not. However, the resultant one had an error. I was planning to upload screenshot now. Does the “alphabet/txt” for the Trie & LM need to be exactly the same as that present in the model. I changed it and there appears to be a problem of gibberish. Can someone clarify if I use existing model (deepspeech -0.3) and my own LM, can the alphabet.txt file be changed?