Toms Bergmanis

dataset.from_files

1

Hi again, I tried to use: `class=dataset.load_dataset_from_files` and `class=dataset.from_files` but that also seems to be out of date as the previous issue I raised:  Is there any up to...

Guided alignments ala Garg et al.

6

Hi all, I just finished reading Sockey 3 paper. Nicely done, congratulations! Have you considered implementing guided alignments[1] in Sockeye 3? It is handy for formatted document translation, non-translatable entity...

Documentation of BERT command line options

4

Marian command line options include: ` --bert-mask-symbol TEXT=[MASK] Masking symbol for BERT masked-LM training` ` --bert-sep-symbol TEXT=[SEP] Sentence separator symbol for BERT next sentence prediction training` ` --bert-class-symbol TEXT=[CLS] Class...

### Bug description Sockeye is training much faster than Marian. I run a 1 data epoch long training on a 4.7M training examples small data set with either framework. To...

Guided alignments in Sockeye.

6

Hi! We finally have time to reopen [work on guided alignments for Sockeye 3](https://github.com/awslabs/sockeye/issues/1054) To recap: guided alignments are handy for formatted document translation, non-translatable entity and placeholder handling, and...

The documentation says that word weights can be real-valued. Real numbers can be negative. This is convenient because outputs of models one could use for scoring often output log-likelihoods, which...