Failure to create the BPE dataset for custom dataset #22
Description
Hi
I am trying to train a style transfer model for a style (i.e., profane vs. civil) that is not supported in the paper. However, when I tried to run the first step as is instructed in the repository
python datasets/dataset2bpe.py --dataset datasets/golbeckwhere datasets/golbeck is a dataset on toxicity comments with required directory structure.
golbeck/
├── dev.label
├── dev.txt
├── test.label
├── test.txt
├── train.label
└── train.txta series of errors on some dependencies are reported.
[...]lib/python3.6/site-packages/torch/include/ATen/core/TensorMethods.h:1417:15: note: no known conversion for argument 1 from ‘<brace-enclosed initializer list>’ to ‘at::TensorList {aka c10::ArrayRef<at::Tensor>}’
error: command 'gcc' failed with exit status 1It seems to me that this should be related to the installation. Here are the full installation commands I used.
conda create --name strap python==3.6.3
conda activate strap
conda install pytorch==1.4.0 torchvision==0.5.0 cudatoolkit=10.1 -c pytorch
pip install -r requirements.txt
pip install -e .
cd fairseq
pip install -e .
# this line is required because otherwise running dataset2bpe.py will report missing dependencies
conda install -c conda-forge hydra-core omegaconfI am wondering how I could resolve this issue. In order to reproduce this error, the data is provided here.