Skip to content

Conversation

@KarelVesely84
Copy link
Contributor

  • there is the 3layer granularity: threads in a binary, mid-level processes
    (summing stats and saving them), top-level processes (-nj).
  • in our cluster it is difficult to get a JOB with too many threads
    (the jobs get stuck in the queue).
  • if the settings are exposed into the top-level script, it will be
    easier to change to setup to make it work well by modifying the top
    level script only...
- there is the 3layer granularity: threads in a binary, mid-level processes (summing stats and saving them), top-level processes (-nj). - in our cluster it is difficult to get a JOB with too many threads (the jobs get stuck in the queue). - if the settings are exposed into the top-level script, it will be easier to change to setup to make it work well by modifying the top level script only...
@KarelVesely84 KarelVesely84 requested a review from danpovey April 18, 2019 13:11

# training chunk-options
chunk_width=140,100,160
chunk_width=140,100,60
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

can you not change the chunk_width? This could affect the results.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I was written in documentation, that the chunk-width should be a list of integers in descending order (try length 140, 100, ...), therefore the 160 did not make sense to me, which is why I changed it. Or am I wrong? What was the intention of using the values '140,100,160' ?

@danpovey danpovey merged commit 286e8af into kaldi-asr:master Apr 23, 2019
@KarelVesely84 KarelVesely84 deleted the wsj_ivector_numjobs branch June 17, 2019 14:11
danpovey pushed a commit to danpovey/kaldi that referenced this pull request Jun 19, 2019
…un script (kaldi-asr#3248) * wsj: exposing i-vector #jobs config to the master script, - there is the 3layer granularity: threads in a binary, mid-level processes (summing stats and saving them), top-level processes (-nj). - in our cluster it is difficult to get a JOB with too many threads (the jobs get stuck in the queue). - if the settings are exposed into the top-level script, it will be easier to change to setup to make it work well by modifying the top level script only...
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

2 participants