File size: 1,031 Bytes
c0d1392 07c73c4 c0d1392 07c73c4 c1433ec c0d1392 c1433ec c0d1392 c1433ec c0d1392 07c73c4 c1433ec |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 |
#!/usr/bin/env bash
#CUDA_VISIBLE_DEVICES="0" python run_speech_recognition_seq2seq.py \
python -m torch.distributed.launch \
--nproc_per_node 8 run_speech_recognition_seq2seq.py \
--dataset_name="librispeech_asr" \
--model_name_or_path="./" \
--dataset_config_name="clean" \
--train_split_name="train.100" \
--eval_split_name="validation" \
--output_dir="./" \
--preprocessing_num_workers="16" \
--length_column_name="input_length" \
--overwrite_output_dir \
--num_train_epochs="5" \
--per_device_train_batch_size="4" \
--per_device_eval_batch_size="4" \
--gradient_accumulation_steps="2" \
--generation_max_length="40" \
--generation_num_beams="1" \
--learning_rate="3e-4" \
--warmup_steps="500" \
--evaluation_strategy="steps" \
--text_column_name="text" \
--save_steps="500" \
--eval_steps="500" \
--logging_steps="1" \
--save_total_limit="1" \
--freeze_feature_extractor \
--gradient_checkpointing \
--fp16 \
--group_by_length \
--predict_with_generate \
--do_lower_case \
--do_eval --do_train
|