|
python run_speech_recognition_ctc.py \ |
|
--dataset_name="mozilla-foundation/common_voice_8_0" \ |
|
--dataset_config_name="nl" \ |
|
--text_column_name="sentence" \ |
|
--length_column_name="input_length" \ |
|
--use_auth_token \ |
|
--model_name_or_path="/workspace/hug_data/xls-r-2b-22-to-16.encoder.ctc48" \ |
|
--tokenizer_name_or_path="./" \ |
|
--output_dir="./output.step1" \ |
|
--overwrite_output_dir \ |
|
--num_train_epochs="5" \ |
|
--per_device_train_batch_size="8" \ |
|
--gradient_accumulation_steps="4" \ |
|
--learning_rate="5e-5" \ |
|
--warmup_steps="2000" \ |
|
--evaluation_strategy="steps" \ |
|
--save_steps="500" \ |
|
--eval_steps="500" \ |
|
--save_total_limit="4" \ |
|
--layerdrop="0.0" \ |
|
--freeze_feature_encoder \ |
|
--gradient_checkpointing \ |
|
--chars_to_ignore , ? . ! - \; \: \" “ % ‘ ” � _ \ |
|
--fp16 \ |
|
--group_by_length \ |
|
--do_train --do_eval |
|
|
|
|
|
|
