mirror of
https://github.com/huggingface/transformers.git
synced 2025-07-31 02:02:21 +06:00
5530299096
* Remove deprecated `evalutate_during_training` * Update src/transformers/training_args_tf.py Co-authored-by: Lysandre Debut <lysandre@huggingface.co> Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
24 lines
906 B
Bash
24 lines
906 B
Bash
export WANDB_PROJECT=distil-marian
|
|
export BS=64
|
|
export GAS=1
|
|
export m=sshleifer/student_marian_en_ro_6_3
|
|
export MAX_LEN=128
|
|
python finetune_trainer.py \
|
|
--tokenizer_name $m --model_name_or_path $m \
|
|
--data_dir $ENRO_DIR \
|
|
--output_dir marian_en_ro_6_3 --overwrite_output_dir \
|
|
--learning_rate=3e-4 \
|
|
--warmup_steps 500 --sortish_sampler \
|
|
--fp16 \
|
|
--gradient_accumulation_steps=$GAS \
|
|
--per_device_train_batch_size=$BS --per_device_eval_batch_size=$BS \
|
|
--freeze_encoder --freeze_embeds \
|
|
--num_train_epochs=6 \
|
|
--save_steps 3000 --eval_steps 3000 \
|
|
--max_source_length $MAX_LEN --max_target_length $MAX_LEN --val_max_target_length $MAX_LEN --test_max_target_length $MAX_LEN \
|
|
--do_train --do_eval --do_predict \
|
|
--evaluation_strategy steps \
|
|
--predict_with_generate --logging_first_step \
|
|
--task translation --label_smoothing 0.1 \
|
|
"$@"
|