This is the official codes for Winner-Take-All Column Row Sampling for Memory Efficient Adaptation of Language Model. FOR REVIEW ONLY, DO NOT DISTRIBUTE
datasets==1.6.2
scikit-learn==0.24.2
tensorboard==2.5.0
matplotlib==3.4.2
transformers==4.6.0
numpy==1.21.1
pip sinstall -v -e .
Run DIVISION on T5 language models:
for dataset in ("rte" "mrpc" "stsb" "cola" "sst2" "qnli" "qqp" "mnli")
do
for seed in (0 1 2)
do
bash scripts/approx_linear.sh 0 $dataset 1 1 1 1 1 1 8 4 division
done
done
Our code is based on the official code of Ladder Site Tuning