machelreid / afromt Goto Github PK

Code for the EMNLP 2021 Paper "AfroMT: Pretraining Strategies and Reproducible Benchmarks for Translation of 8 African Languages" by Machel Reid, Junjie Hu, Graham Neubig, Yutaka Matsuo

Shell 0.13% Python 97.38% C++ 0.58% Cuda 1.32% Cython 0.45% Lua 0.14%

nlp machine-learning low-resource-machine-translation

afromt's People

Contributors

Stargazers

Watchers

Forkers

andrews2017 ghw2016z

afromt's Issues

Data Issues

Good Morning dear Machel, Please can you give me the access to the data that you use for pre-training stage and also the corresponding task use during pre-training.
Best Regards

sacrebleu warning

Please can I have some explanation about this warning???

Training issue

I'm having the below error during the training process.

Below picture is the content of training script

Hello dear Machelreidm, I followed the instructions in the afromt repository to the end, but the evaluation returns 0 for both metrics.
cat tgt | sacremoses detokenize > tgt.tmp 2> /dev/null
cat hyp | sacremoses detokenize > hyp.tmp 2> /dev/null
lThe two lines of code above are the cause because after executing these two lines the tgt.tmp and hyp.tmp files are empty.
ref: https://github.com/machelreid/afromt/blob/main/eval_bleu_chrf.sh#L5
I am hereby asking for your suggestions.
Cordially

Recommend Projects