Python Version: Python>=3.6
Package Requirements: torch>=1.4.0 tensorboardX numpy>=1.19.0
Before running the scripts, please install fairseq dependencies by:
pip install --editable .
- Step 1: Combine sentences to create context-aware information:
mkdir exp_mbart
k=3: bash exp_gtrans/run-all.sh prepare-mbart exp_mbart 3
k=5: bash exp_gtrans/run-all.sh prepare-mbart exp_mbart 5
k=7: bash exp_gtrans/run-all.sh prepare-mbart exp_mbart 7
- Step 2: Prepare data:
bash exp_gtrans/run-all.sh prepare-mbart exp_mbart
- Step 3: Train model:
CUDA_VISIBLE_DEVICES=0,1,2,3 bash exp_gtrans/run-all.sh run-mbart train exp_mbart
- Step 4: Evaluate model:
bash exp_gtrans/run-all.sh run-mbart test exp_mbart
Some code are borrowed from G-Transformer. Thanks for their work.