François Hernandez

Hi there, Thanks for the nice papers and implementation! As part of a few set of experiments I'm trying to run this on a custom dataset with around 1500 recordings,...

If a cluster has some quite frequent cronjobs, there are a lot of not necessarily useful notifications. Is there a a way to ignore the CronJob related pods in the...

Upgrade torchtext to more recent releases

5

This warning is raised when using torchtext 0.8.0: ``` warnings.warn('Example class will be retired soon and moved to torchtext.legacy. Please see the most recent release notes for further information.', UserWarning)...

Improve CTranslate2 wrapping in translation_server

5

https://forum.opennmt.net/t/ctranslate2-on-opennmt-py-server/4175/8

Deprecated behavior from torch 1.7

1

The following warning is raised when inferring with torch 1.7.0: ``` /usr/local/lib/python3.6/dist-packages/onmt/translate/beam_search.py:209: UserWarning: An output with one or more elements was resized since it had shape [40], which does not...

Basic idea: give an additional file when preprocessing, containing sentence weights. These weights will be stored in the torchtext `Example`s and will be used to weight the loss in `_compute_loss`....

[WIP] Keep track of loaded shards

1

This would allow to restart from (approximately) the same point in the data when continuing some training, instead of restarting from shard 0 each time.

[WIP v1 - deprecated] Add learning of target features

6

This feature has been asked about quite a lot, and might be useful in several tasks, so here it is! ### Done - [x] create a `Generator` class to handle...

[WIP] The Missing Ingredient in Zero-Shot Neural Machine Translation

2

This PR intends to add an implementation of the cosine similarity alignment loss introduced as a regularization term in [_The Missing Ingredient in Zero-Shot Neural Machine Translation_](https://arxiv.org/abs/1903.07091).

[WIP] LAMB optimizer

1

[DO NOT MERGE] This is a WIP on implementing LAMB optimizer from BERT. It apparently allows to scale training on huge batches. There are some ambiguities : different algorithms between...