|
export WANDB_ENTITY=cahya |
|
export WANDB_LOG_MODEL=true |
|
export WANDB_PROJECT=xlsr-turkish |
|
|
|
python run_speech_recognition_ctc.py \ |
|
--dataset_name="mozilla-foundation/common_voice_7_0" \ |
|
--model_name_or_path="cahya/wav2vec2-base-turkish-artificial" \ |
|
--dataset_config_name="tr" \ |
|
--output_dir="./" \ |
|
--overwrite_output_dir \ |
|
--num_train_epochs="100" \ |
|
--per_device_train_batch_size="128" \ |
|
--per_device_eval_batch_size="8" \ |
|
--gradient_accumulation_steps="4" \ |
|
--learning_rate="3e-4" \ |
|
--warmup_steps="100" \ |
|
--length_column_name="input_length" \ |
|
--evaluation_strategy="steps" \ |
|
--text_column_name="sentence" \ |
|
--save_steps="200" \ |
|
--eval_steps="200" \ |
|
--logging_steps="200" \ |
|
--layerdrop="0.0" \ |
|
--activation_dropout="0.1" \ |
|
--save_total_limit="3" \ |
|
--freeze_feature_encoder \ |
|
--feat_proj_dropout="0.0" \ |
|
--mask_time_prob="0.75" \ |
|
--mask_time_length="10" \ |
|
--mask_feature_prob="0.25" \ |
|
--mask_feature_length="64" \ |
|
--gradient_checkpointing \ |
|
--use_auth_token \ |
|
--fp16=true \ |
|
--group_by_length \ |
|
--do_train=true \ |
|
--do_eval=true \ |
|
--push_to_hub \ |
|
--chars_to_ignore , ? . ! \; \: \"\" \% \' \" \' \' \` … \’ » « \‘ '“' '”' � é û \( \) |
|
|