Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
Olivier Bondu
@obondu:matrix.org
[m]
Alix Chagué: thanks for the model on Zenodo ! I just tried it and it works like a charm :)
Louis-Fiacre
@Louis-Fiacre
Hello, I have a question about Cuda on Docker. The documentation does not give much information, can I use the latest version of Cuda or do I need a specific version ?
5 replies
Daniel Stoekl
@dstoekl
@Lauxley_gitlab can I write a msg on the list that people can return working on msia? afaics all is back to normal. Many sincere thanks to @mittagessen, yourself, @babadie_gitlab and Stéfane and all others involved in the detective work to resolve the issue quicker than feared!
3 replies
Daniel Stoekl
@dstoekl
Just trained a model on top of another one. Process finished. No accuracy shown.
Same training on vietnamica crashed.
Daniel Stoekl
@dstoekl
current training does not give any accuracy either and continues for an eternity despite a very small training set.
@Lauxley_gitlab file an issue?
Robin Tissot
@Lauxley_gitlab
ermm nothing changes in that regard though
Robin Tissot
@Lauxley_gitlab
from scratch?
Daniel Stoekl
@dstoekl
never training from scratch
Robin Tissot
@Lauxley_gitlab
RuntimeError: Found no NVIDIA driver on your system. Please check that you have an NVIDIA GPU and installed a driver from http://www.nvidia.com/Download/index.aspx
looks like something broke in the os update
Robin Tissot
@Lauxley_gitlab
6 replies
Bastien Abadie
@babadie_gitlab

@Lauxley_gitlab Hey :wave:

When do you plan to build the next release, to showcase search v1 ?

Itay Zandbank
@zmbq_gitlab
@Lauxley_gitlab , in the commit scripta/escriptorium@874e8a2 , you changed the task signal so that it checks sender.name and not sender. This fails, since sender is a string. Should there be a change in the requirements file, so that a one of the celery packages is updated and the sender becomes an object?
6 replies
Itay Zandbank
@zmbq_gitlab

@Lauxley_gitlab , I think I found the issue with moving multiple lines at once. The moving code assumes the order of a line only changes once. Suppose we have the lines ABCDEF (A at position 0, B at position 1 and so on), and we want to reverse the order - to FEDCBA. The moving code knows it needs to move line A from 0 to 5. So it does so, and now the order is BCDEFA. Now it moves line B from 1 to 4. However, B is no longer at position 1, it is at position 0. So the order now becomes BDEFCA. This inconsistency continues.

I may be able to figure out a way to fix this, however - it will be rather messy code, and also quite inefficient (as the django-ordered-model code updates O(N) records when moving one record, so we will get O(N^2) updates when applying a full permutation). Not using django-ordered-models for applying a permutation will result in much cleaner and also faster code.

5 replies
Daniel Stoekl
@dstoekl
there are 3 tasks noted as running on msia two of which do not make sense because they were launched many days ago. can someone tell me what is happening? we need the taskmonitor to specify whether its training, seg, reco or whatever I think
2 replies
Thibault Lavigne
@tlavigne_gitlab
@dstoekl @Lauxley_gitlab v0.12.4c is deployed on msia dev
2 replies
Daniel Stoekl
@dstoekl
we really need the bugrepairs for the line joining before the next redeployment
they have a huge impact
will write a message to Itay
Daniel Stoekl
@dstoekl
I need to add diacritics to Hebrew letters in the transcription modal and elsewhere but the display always puts them AFTER the character instead of above. Anyone having a solution?
e.g.
א
֯
should be
א֯
but in eScriptorium it looks like
֯ א
mittagessen
@mittagessen
That's a problem with the font and not escriptorium I'd wager. If there isn't a pre-composed glyph for the character it will look like the diacritic is offset to the left.
Daniel Stoekl
@dstoekl
then why can gitter display it correctly but eScr cannot?
mittagessen
@mittagessen
In fact the diacritizied letter you typed looks 'wrong' on my end because my font doesn't have a glyph for that either.
different font I guess.
Daniel Stoekl
@dstoekl
its both in firefox
mittagessen
@mittagessen
yes but you can set preferred fonts on each page.
Daniel Stoekl
@dstoekl
how ?
mittagessen
@mittagessen
CSS
Daniel Stoekl
@dstoekl
so you mean @Lauxley_gitlab can set up preferred fonts on my page
because I did not change anything
mittagessen
@mittagessen
it is a fundamentally unsolvable problem because you don't know which precomposed glyphs are available on each system.
there's a build-in fallback mechanism but it can fail
just by setting a different font size/weight you can get different behavior in the font selection process.
Daniel Stoekl
@dstoekl
I will see what I can do with stylus and the firefox prefs
mittagessen
@mittagessen
but it isn't easily introspectable at all because that whole algorithm is deeply buried in the font shaper.
my suggestion is to just accept it. It doesn't change anything in the recognition or export, only makes the glyph look uglier (on some systems).
Daniel Stoekl
@dstoekl
it makes it much harder to read and check for wrong spaces
which is the most current error
anyhow I already filed enough issues today
I dont want him to get mad at me
Daniel Stoekl
@dstoekl
@babadie_gitlab the search index on the dev has not been built so we cannot test.