Efficient Web Scale Abstractive summarization leveraging model asymmetry, scaling laws, and pseudo labeling

python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/small/ asy/xsum/small-5-6 small 5 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/small/ asy/xsum/small-4-6 small 4 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/small/ asy/xsum/small-3-6 small 3 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/small/ asy/xsum/small-2-6 small 2 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/small/ asy/xsum/small-1-6 small 1 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/small/ asy/xsum/small-6-5 small 6 5 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/small/ asy/xsum/small-6-4 small 6 4 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/small/ asy/xsum/small-6-3 small 6 3 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/small/ asy/xsum/small-6-2 small 6 2 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/small/ asy/xsum/small-6-1 small 6 1 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/small/ asy/xsum/small-5-5 small 5 5 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/small/ asy/xsum/small-4-4 small 4 4 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/small/ asy/xsum/small-3-3 small 3 3 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/small/ asy/xsum/small-2-2 small 2 2 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/small/ asy/xsum/small-1-1 small 1 1 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/base/ asy/xsum/base-5-6 base 5 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/base/ asy/xsum/base-4-6 base 4 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/base/ asy/xsum/base-3-6 base 3 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/base/ asy/xsum/base-2-6 base 2 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/base/ asy/xsum/base-1-6 base 1 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/base/ asy/xsum/base-6-5 base 6 5 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/base/ asy/xsum/base-6-4 base 6 4 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/base/ asy/xsum/base-6-3 base 6 3 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/base/ asy/xsum/base-6-2 base 6 2 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/base/ asy/xsum/base-6-1 base 6 1 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/base/ asy/xsum/base-5-5 base 5 5 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/base/ asy/xsum/base-4-4 base 4 4 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/base/ asy/xsum/base-3-3 base 3 3 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/base/ asy/xsum/base-2-2 base 2 2 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/xsum/base/ asy/xsum/base-1-1 base 1 1 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/small/ asy/cnndm/small-5-6 small 5 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/small/ asy/cnndm/small-4-6 small 4 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/small/ asy/cnndm/small-3-6 small 3 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/small/ asy/cnndm/small-2-6 small 2 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/small/ asy/cnndm/small-1-6 small 1 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/small/ asy/cnndm/small-6-5 small 6 5 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/small/ asy/cnndm/small-6-4 small 6 4 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/small/ asy/cnndm/small-6-3 small 6 3 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/small/ asy/cnndm/small-6-2 small 6 2 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/small/ asy/cnndm/small-6-1 small 6 1 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/small/ asy/cnndm/small-5-5 small 5 5 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/small/ asy/cnndm/small-4-4 small 4 4 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/small/ asy/cnndm/small-3-3 small 3 3 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/small/ asy/cnndm/small-2-2 small 2 2 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/small/ asy/cnndm/small-1-1 small 1 1 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/base/ asy/cnndm/base-5-6 base 5 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/base/ asy/cnndm/base-4-6 base 4 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/base/ asy/cnndm/base-3-6 base 3 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/base/ asy/cnndm/base-2-6 base 2 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/base/ asy/cnndm/base-1-6 base 1 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/base/ asy/cnndm/base-6-5 base 6 5 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/base/ asy/cnndm/base-6-4 base 6 4 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/base/ asy/cnndm/base-6-3 base 6 3 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/base/ asy/cnndm/base-6-2 base 6 2 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/base/ asy/cnndm/base-6-1 base 6 1 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/base/ asy/cnndm/base-5-5 base 5 5 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/base/ asy/cnndm/base-4-4 base 4 4 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/base/ asy/cnndm/base-3-3 base 3 3 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/base/ asy/cnndm/base-2-2 base 2 2 python Efficient-Web-Scale-Absractive-Summarization/prune.py asy/cnndm/base/ asy/cnndm/base-1-1 base 1 1

python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-cnndm cnn/large-1-1 large 1 1 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-cnndm cnn/large-2-2 large 2 2 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-cnndm cnn/large-3-3 large 3 3 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-cnndm cnn/large-4-4 large 4 4 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-cnndm cnn/large-5-5 large 5 5 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-cnndm cnn/large-5-6 large 5 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-cnndm cnn/large-4-6 large 4 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-cnndm cnn/large-3-6 large 3 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-cnndm cnn/large-2-6 large 2 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-cnndm cnn/large-1-6 large 1 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-cnndm cnn/large-6-5 large 6 5 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-cnndm cnn/large-6-4 large 6 4 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-cnndm cnn/large-6-3 large 6 3 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-cnndm cnn/large-6-2 large 6 2 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-cnndm cnn/large-6-1 large 6 1

python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-xsum x/large-1-1 large 1 1 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-xsum x/large-2-2 large 2 2 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-xsum x/large-3-3 large 3 3 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-xsum x/large-4-4 large 4 4 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-xsum x/large-5-5 large 5 5 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-xsum x/large-5-6 large 5 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-xsum x/large-4-6 large 4 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-xsum x/large-3-6 large 3 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-xsum x/large-2-6 large 2 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-xsum x/large-1-6 large 1 6 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-xsum x/large-6-5 large 6 5 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-xsum x/large-6-4 large 6 4 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-xsum x/large-6-3 large 6 3 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-xsum x/large-6-2 large 6 2 python Efficient-Web-Scale-Absractive-Summarization/prune.py spacemanidol/flan-t5-large-xsum x/large-6-1 large 6 1

CUDA_VISIBLE_DEVICES=1 python run_summarization.py --model_name_or_path asy/cnndm/small-1-1 --do_train --do_eval --dataset_name cnn_dailymail --dataset_config "3.0.0" --source_prefix summarize: --per_device_train_batch_size=4 --gradient_accumulation_steps 16 --learning_rate 1e-4 --output_dir asy/cnndm/small-1-1-t --lr_scheduler_type constant --per_device_eval_batch_size=4 --overwrite_output_dir --predict_with_generate --weight_decay 0.01 --save_total_limit 1

CUDA_VISIBLE_DEVICES=0 python run_summarization.py --do_train --do_eval --dataset_name cnn_dailymail --dataset_config 3.0.0 --ummarize: --per_device_train_batch_size=2 --gradient_accumulation_steps 132 --learning_rate 1e-4 --lr_scheduler_type constant --save_total_limit 1 --per_device_eval_batch_size=4 --overwrite_output_dir --predict_with_generate --fp16 --weight_decay 0.01 --model_name_or_path asy/cnndm/small-6-5 --output_dir asy/cnndm/small-6-5-t/