Kelly Marchisio

In the readme it says, "In addition, you can also evaluate the model in the same settings as in our paper using the evaluate.py script." -- note that there is...

I'm running the below command, however my results are not deterministic. Is there a way to do so? (I'm debugging the pipeline, so would like a fully reproducible run) ```...

Excessive memory usage during ensemble decoding

4

We're ensemble decoding with 4 models using this command: ``` $MARIAN/build/marian-decoder \ -c $configs \ -m $models -d $GPU \ --mini-batch 16 --maxi-batch 100 --maxi-batch-sort src -w 6500 \ --n-best...

## 🐛 Bug When we specify nwords=N, in tasks.build_dictionary [here](https://github.com/facebookresearch/fairseq/blob/b4001184f49ed0e20d619b54bb3d43088fabf990/fairseq/tasks/fairseq_task.py#L97), if nwords is less than self.symbols, the resulting dictionary is size len(self.symbols) (code [snippet](https://github.com/facebookresearch/fairseq/blob/b4001184f49ed0e20d619b54bb3d43088fabf990/fairseq/data/dictionary.py#L174)). Perhaps this is intended behavior, but...

I added some sequences of repeated characters as user defined tokens to a Unigram model. Now when tokenizing with sampling, I get unexpected behavior as I increase the nbest size....