-
Hello,
I noticed the BertForSeq2SeqDecoder to be slow on CPU and this is mainly due to the while looping inside the forward method, where basically it iterates N times where N is the difference be…
-
The FastSpeech paper said, ‘We feed the text and speech pairs in the training set to the model again to obtain the encoder-decoder attention alignments, which are used to train the duration predictor’…
nanma updated
4 years ago
-
Hi Haijun, thanks for sharing your code, it is great implementation.
As I go through your code, I found you did not adopt teacher forcing to train the RNN decoder. I would like to know, if it is poss…
-
Hi, i just fast implemented FastSpeech2 to check the contribution of F0 embedding and Energy embedding. See PR #45 for detail.
-
Hi,
Thank you for publishing nonautoregressive mt code. I have tried it following your introduction. The training loss decreases normally. But I got bleu 0.0 in newstest2014 en2de.
Data is prepr…
-
Hi, nice work you've done! I was wondering if you had any visualisation of trained vq embedding space? I've been trying to reproduce the paper as well, but the model seems not to use the global condit…
ghost updated
4 years ago
-
Hi,
I am trying to find an easy way to add more layers to the autoregressive flow network via `tfp.bijectors.masked_autoregressive_default_template()`. I am not sure I understand what network arch…
-
in this commit: [https://github.com/statsmodels/statsmodels/commit/d03474da1aae5dac54c2b4441311d01517cd2567](url) the sarimax model was changed to only warn on non-invertible/stationary start_params a…
-
#### Describe the bug
SARIMAX does not work with directly specifying lags. Appears to have a bug in starting value code.
This is using latest dev (Nov 12, 2019)
#### Code Sample, a copy-pasta…
-
when I tried this command : cat testing_files/output_for_multitask_bpe.txt | python ~/fairseq/interactive.py ~/fairseq/data-bin/eli5_data --path multitask_checkpoint.pt --task translation --batch-siz…