MultiESC

Running Scripts for Muffin (MultiESC)

Training Base Model

Refer to the original codes for more details.

Training Muffin Model

Run:

python mitigation_MultiESC.py --data_type=8 --model_name_or_path=./final_output/whlookahead_generate --learning_rate=3e-5 --lr2=1e-4 --with_cause --with_strategy --lookahead --model_type=1 --candidate_num=10 --per_device_train_batch_size=8 --gradient_accumulation_steps=2 --warmup_steps=1200