Open dsli208 opened 2 years ago
Adding the option "-lang_class 0 1" should solve the issue. Let me know if you've got any further questions.
When I try this, a new issue will arise when I try and run train.py
:
RuntimeError: CUDA out of memory. Tried to allocate 7.06 GiB (GPU 0; 11.91 GiB total capacity; 3.55 GiB already allocated; 6.99 GiB free; 3.99 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
That's an OOM error. You may either use a GPU with larger memories or use multiple GPUs (e.g. by setting CUDA_VISIBLE_DEVICES=0,1)
issue still seems to persist after i try this
In that case, you may use more GPUs (CUDA_VISIBLE_DEVICES=0,1,2). Alternatively, before you apply the following preprocess:
python filter_long_sent.py -src ${de} -tgt ${en}
you may reduce the cut-off threshold in "filter_long_sent.py"
if len(src[i].split())<=80 and len(tgt[i].split())<=80:
from 80 to 70/60 to remove long sentences, and run python preprocess.py again. Let me know if it doesn't still go well. I appreciate your patience.
I'm trying to follow the instructions for reproducing the DE-EN word alignment experiment, but I am hitting a snag when trying to run
train.py
:Is there an issue with the file that I should be aware of, or is there a quick fix for such an issue? I ran the
--help
flag and I'm apparently missing aLANG_CLASS
, is that provided as part of reproducing said experiment?