Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
  • May 05 2021 16:41
    guillaumekln transferred #580
  • May 05 2021 16:22
    hangcao1004 opened #580
  • Apr 16 2021 02:57
    raymondhs closed #400
  • May 29 2020 07:14
    guillaumekln closed #579
  • May 29 2020 07:14
    guillaumekln commented #579
  • May 28 2020 20:37
    Roweida-Mohammed edited #579
  • May 28 2020 20:36
    Roweida-Mohammed opened #579
  • Feb 19 2020 16:08
    codecov-io commented #578
  • Feb 19 2020 16:08

    guillaumekln on master

    Updating the intel-mkl URL. (#5… (compare)

  • Feb 19 2020 16:08
    guillaumekln closed #578
  • Feb 19 2020 16:08
    guillaumekln commented #578
  • Feb 19 2020 15:59
    arturgontijo opened #578
  • Feb 12 2020 17:58
    melindaloubser1 closed #553
  • Feb 12 2020 17:58
    melindaloubser1 commented #553
  • Dec 13 2019 08:43
    guillaumekln transferred #574
  • Dec 13 2019 08:43
    guillaumekln transferred #577
  • Dec 13 2019 08:14
    tkngoutham edited #577
  • Dec 13 2019 08:13
    tkngoutham opened #577
  • Dec 13 2019 06:37
    tkngoutham commented #574
  • Oct 09 2019 11:36

    guillaumekln on master

    Add CTranslate2 Change project cards title (compare)

Ratish Puduppully
@ratishsp
The perplexity of concat is only slightly higher than that of general
Jean Senellart
@jsenellart
Hi @ratishsp, I found the same although the difference reduces for larger model.
Jean Senellart
@jsenellart
@vince62s - first unsuccessful try tonight CUDA 9 / torch. There some complaints on torch gitter about the same. Did you try?
Vincent Nguyen
@vince62s
no I read similar stuff, but on the other hand TF 1.5 is now distributed with binary compiled on cuda 9. So would be good to find a solution.
how do people do with new V100 then ?
Jean Senellart
@jsenellart
I am trying to narrow down - it seems to be coming from incompatibility with some other library.
Vincent Nguyen
@vince62s
I guess you tried this export TORCH_NVCC_FLAGS="-D__CUDA_NO_HALF_OPERATORS__"right ?
Jean Senellart
@jsenellart
yes - it was not the point
Ratish Puduppully
@ratishsp
Hi, how do we disable dropout during testing?
Guillaume Klein
@guillaumekln
Hello, it is disabled automatically.
Ratish Puduppully
@ratishsp
Thanks @guillaumekln.
chiting765
@chiting765
Hi~ I have a question about joiner annotate. There is a -tok_joiner_annotate option at learn_bpe.lua and a -joiner_annotate option at tokenize.lua. Do I need to turn on both to have the joiner in the tokenized files?
Jean Senellart
@jsenellart
@chiting765 - I think the one in learn_bpe is not important. It is just coming with all tokenization options
it does not change the bpe model
so the only one important is the one in tokenize.lua
chiting765
@chiting765
@jsenellart OK~ Thanks! The first time I tired the bpe model for EN to ES translation, I got target word like "Kilogram o" which supposed to be "Kilogramo". I think it is because I didn't add the joiner annotator. I am training a new one with the joiner annotator and hopefully this time I will get the word correctly
chiting765
@chiting765
Hi~ So I tried bpe model with joiner annotator, it did give me correct target words like "Kilogramo". The validation ppl is also lowered from 2.5 to 2.2. However, the BLUE score is lowered too, I am not sure why
BLEU score
Jean Senellart
@jsenellart
is your bleu score calculated on tokenized or detokenized?
chiting765
@chiting765
detokenized
I will try to calculate the BLEU of space tokenized target file
chiting765
@chiting765
I calculated the BLEU of the space tokenized candidate file against the space tokenized reference file. The BLEU did improve from 54 to 55. However, without the BPE, the best BLEU I got for the same file is 65
I wonder maybe I should try a different BPE model other the aggressive one?
What kind of BPE model do you use for EN to ES translation or to similar languages?
Jean Senellart
@jsenellart
your result is unlikely except if you have a very small vocabulary/narrow domain. A BLEU of 65 is suspicious
there are almost no cases where we found BPE hurting the translation
chiting765
@chiting765
It is a pretty narrow domain, the vocabulary is not very small, it has about 40K - 50K vocabulary in total without bpe
and the BLEU score is for the whole test data
zeng
@xjtu-zeng
Hi everyone. I have a question about the StdRNNDecoder, why the rnn and attn can be seperated? The next hidden state needs the context computed by attn. I am confused
@jsenellart
Vincent Nguyen
@vince62s
Just in case one did not notice, but there is a huge performance difference between Cuda 8.0.61 and 8.0.61 patch 2 for the best (I saw about 25% difference)
Jean Senellart
@jsenellart
Registration for first OpenNMT workshop is open! Check here: http://workshop-paris-2018.opennmt.net :)...
ykasimov
@ykasimov
Hi. There is no support for copy attention yet in Python version, right?
Konstantin Glushak
@gsoul
Which of python versions did you mean?
ykasimov
@ykasimov
do you mean python version?
Konstantin Glushak
@gsoul
no, OpenNMT version: OpenNMT-py or OpenNMT-tf?
ykasimov
@ykasimov
ah, sorry. OpenNMT-py. Forgot that there is tf version
Konstantin Glushak
@gsoul
I’m not sure, but perhaps it’s better to ask this question in OpenNMT-py channel? https://gitter.im/OpenNMT/OpenNMT-py
ykasimov
@ykasimov
Thanks.
Konstantin Glushak
@gsoul
np
Ratish Puduppully
@ratishsp

In GlobalAttention.lua, we have the following lines of code
local softmaxAttn = nn.SoftMax() softmaxAttn.name = 'softmaxAttn'

Why don't we set softmaxAttn as an output of nn.gModule like return nn.gModule(inputs, {contextOutput, softmaxAttn(attn)})

Jean Senellart
@jsenellart
what for?
it is not used later
but we name it, so that we can find it by traversing the graph
Ratish Puduppully
@ratishsp
Ok. I was trying to understand the design difference between the two: when should we set it as an output of nn.gModule and when should we not.
Jean Senellart
@jsenellart
the gModule is very powerful but also very complicated - you can not easily tweak it
Ratish Puduppully
@ratishsp
Ok.
I guess if it is output of nn.gModule, then we should manage its backpropagation with gradients too.
Jean Senellart
@jsenellart
yes exactly - for the attention, we are just accessing for visualization of the state
Ratish Puduppully
@ratishsp
Thanks @jsenellart for the details.