-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathsubmit.sh
19 lines (16 loc) · 898 Bytes
/
submit.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
# NSML Training examples
nsml run -g 5 -d {nsml_dataset} -e run.py --esm {your_id} -a '--model_name_or_path hyunwoongko/kobart
--do_train True
--num_train_epochs 6
--dataset_name {Huggingface_dataset_hub_full_name}
--source_prefix "요약: "
--output_dir ./bart-v2-x-fp32
--overwrite_output_dir True
--per_device_train_batch_size 22
--gradient_accumulation_steps 8
--text_column input
--summary_column output
--learning_rate 1e-5
--seed 42'
nsml run -g 4 -d {nsml_dataset} -e main.py --esm {your_id} -a "--epochs 8 --max_target_len 256 --lr 1e-4 --model paust/pko-t5-large --fp16 True --batch 2 --gradient_accum 8 --overwrite_cache True"
nsml run -d {nsml_dataset} -e main.py --esm {your_id} -a "--epochs 3 --model paust/pko-t5-large --lr 1e-4 --load_session {full_nsml_session_name} --fp16 True --batch 12 --eval_batch 52 --gradient_accum 8"