|
#!/bin/bash |
|
|
|
export LC_ALL=C.UTF-8 |
|
export LANG=C.UTF-8 |
|
|
|
export OUTPUT_DIR=/to/our/path |
|
export MODEL_NAME_OR_PATH=t5-base |
|
export NUM_BEAMS=3 |
|
|
|
export TRAIN_FILE=/to/../train.csv |
|
export VALIDATION_FILE=/to/../dev.csv |
|
export TEST_FILE=/to/../test.csv |
|
export TEXT_COLUMN=inputs |
|
export TARGET_COLUMN=targets |
|
export MAX_SOURCE_LENGTH=128 |
|
export MAX_TARGET_LENGTH=1024 |
|
export SOURCE_PREFIX=ingredients |
|
|
|
export PER_DEVICE_TRAIN_BATCH_SIZE=8 |
|
export PER_DEVICE_EVAL_BATCH_SIZE=8 |
|
export GRADIENT_ACCUMULATION_STEPS=2 |
|
export NUM_TRAIN_EPOCHS=3.0 |
|
export LEARNING_RATE=5e-4 |
|
export WARMUP_STEPS=5000 |
|
|
|
python run_ed_recipe_nlg.py \ |
|
--output_dir="$OUTPUT_DIR" \ |
|
--train_file="$TRAIN_FILE" \ |
|
--validation_file="$VALIDATION_FILE" \ |
|
--test_file="$TEST_FILE" \ |
|
--text_column="$TEXT_COLUMN" \ |
|
--target_column="$TARGET_COLUMN" \ |
|
--source_prefix="$SOURCE_PREFIX: " \ |
|
--max_source_length="$MAX_SOURCE_LENGTH" \ |
|
--max_target_length="$MAX_TARGET_LENGTH" \ |
|
--model_name_or_path="$MODEL_NAME_OR_PATH" \ |
|
--extra_tokens="" \ |
|
--special_tokens="<sep>,<items>" \ |
|
--per_device_train_batch_size=$PER_DEVICE_TRAIN_BATCH_SIZE \ |
|
--per_device_eval_batch_size=$PER_DEVICE_EVAL_BATCH_SIZE \ |
|
--gradient_accumulation_steps=$GRADIENT_ACCUMULATION_STEPS \ |
|
--num_train_epochs=$NUM_TRAIN_EPOCHS \ |
|
--learning_rate=$LEARNING_RATE \ |
|
--warmup_steps=$WARMUP_STEPS \ |
|
--preprocessing_num_workers=4 \ |
|
--prediction_debug \ |
|
--do_train \ |
|
--do_eval \ |
|
--do_predict \ |
|
--overwrite_output_dir \ |
|
--predict_with_generate |
