正确的版本
nvidia-smi
cd $SLURM_SUBMIT_DIR
NAME=e30-gen-py-raw
LOSS_FILENAME='loss/'$NAME'.csv'
LOG_FILENAME='log/'$NAME'.log'
TRAIN_FILE='py-raw/encoder-train.json.seq2seq'
DEV_FILE='py-raw/encoder-dev.json.seq2seq'
TEST_FILE='py-raw/encoder-test.json.seq2seq'
MODEL_NAME_OR_PATH='/project/ruanxiaoming/codes/models/codet5-base'
python3 run.py --do_train --do_eval --do_test --model_name $MODEL_NAME_OR_PATH --train_filename $TRAIN_FILE --dev_filename $DEV_FILE --test_filename $TEST_FILE --loss_filename $LOSS_FILENAME --num_train_epochs 30 --lang bash --max_source_length 256 --max_target_length 128 --train_batch_size 6 --eval_batch_size 6 --log_name=$NAME --output_dir=$NAME
bash evaluate-py.sh $NAME 1
报错经过
- 变量名小写报错(这个待考究,bash变量区分大小写,但是小写应该也行)
train_file='py-raw/encoder-train.json.seq2seq'
/tmp/slurmd/job2426007/slurm_script: line 16: TRAIN_FILE: command not found
- 等号左右有空格报错
train_file = 'py-raw/encoder-train.json.seq2seq'
原因