2023-04-06 14:42:20 +00:00
|
|
|
PRE_SEQ_LEN=128
|
|
|
|
|
LR=2e-2
|
2023-03-31 02:43:55 +00:00
|
|
|
|
|
|
|
|
CUDA_VISIBLE_DEVICES=0 python3 main.py \
|
|
|
|
|
--do_train \
|
|
|
|
|
--train_file AdvertiseGen/train.json \
|
|
|
|
|
--validation_file AdvertiseGen/dev.json \
|
|
|
|
|
--prompt_column content \
|
|
|
|
|
--response_column summary \
|
|
|
|
|
--overwrite_cache \
|
2023-03-31 03:30:36 +00:00
|
|
|
--model_name_or_path THUDM/chatglm-6b \
|
2023-03-31 12:15:35 +00:00
|
|
|
--output_dir output/adgen-chatglm-6b-pt-$PRE_SEQ_LEN-$LR \
|
2023-03-31 02:43:55 +00:00
|
|
|
--overwrite_output_dir \
|
|
|
|
|
--max_source_length 64 \
|
|
|
|
|
--max_target_length 64 \
|
2023-03-31 03:28:13 +00:00
|
|
|
--per_device_train_batch_size 1 \
|
2023-03-31 02:43:55 +00:00
|
|
|
--per_device_eval_batch_size 1 \
|
2023-03-31 03:28:13 +00:00
|
|
|
--gradient_accumulation_steps 16 \
|
2023-03-31 02:43:55 +00:00
|
|
|
--predict_with_generate \
|
|
|
|
|
--max_steps 3000 \
|
|
|
|
|
--logging_steps 10 \
|
|
|
|
|
--save_steps 1000 \
|
|
|
|
|
--learning_rate $LR \
|
|
|
|
|
--pre_seq_len $PRE_SEQ_LEN \
|
|
|
|
|
--quantization_bit 4
|
|
|
|
|
|
