Open guneemwelloeux opened 4 years ago
Thank you for pointing out the segmentation fault. I'm not sure why we get it. The following should work on macOS:
pip install pytext-nlp
pytext train < src/resources/config/ner.json
and the segmentation fault should go awayI opened an issue with the PyText team.
The segmentation fault may be related to the version mismatch between pytext and onnx/torch.
If Anaconda3 by itself does not help, you may have to upgrade onnx and torch:
conda install onnx -c conda-forge
conda install pytorch torchvision -c pytorch
Same issue on Ubuntu 20.04 LTS, Python 3.8.3
On MacOS Mojave and Catalina, training the NER models results in a segfault during the training of epoch 1. I'm using the command as advertised
pytext train < src/resources/config/ner.json
, from the root of the repo. The training is running in CPU mode only.I created a dockerfile to run the same operation, and that one doesn't fail.
Below is the backtrace of the core dump, in case it helps (kept only the frames involving other libraries than python itself, there's an extra 81 frames)