Awesome

Setup

1. Download the code

2. Install modified transformers

cd transformers; pip install .; cd ..

3. Download pre-trained files

we use the pre-trained files provided by our baseline. please refer to the code of the paper "Continual Few-shot Relation Learning via Embedding Space Regularization and Data Augmentation" for details.

4. Run script

python fewrel_5shot_mem_1.2dw.py

5. Attention: The problems you may encounter.

transformer

pip uninstall transformers
pip install transformers==4.5.0

wordninja

pip install wordninja

faiss

conda install -c pytorch faiss-gpu cudatoolkit=11.0

faiss wrong (python3.9 -> 3.8)

strings /usr/lib/x86_64-linux-gnu/libstdc++.so.6 | grep GLIBCXX
unzip libstdc.so_.6.0.26.zip
sudo mv libstdc++.so.6.0.26  /usr/lib/x86_64-linux-gnu/
cd /usr/lib/x86_64-linux-gnu/
sudo rm libstdc++.so.6
ln libstdc++.so.6.0.26 libstdc++.so.6
strings /usr/lib/x86_64-linux-gnu/libstdc++.so.6 | grep GLIBCXX

version

/root/miniconda3/envs/myconda/lib/python3.8/site-packages/transformers/dependency_versions_table.py "tokenizers": "tokenizers>=0.10.1",