#!/usr/bin/env bash

python run_speech_recognition_ctc.py \
	--dataset_name="mozilla-foundation/common_voice_7_0" \
	--model_name_or_path="facebook/wav2vec2-xls-r-1b" \
	--dataset_config_name="uk" \
	--output_dir="./cv-uk-ft" \
	--num_train_epochs="100" \
	--per_device_train_batch_size="8" \
	--per_device_eval_batch_size="8" \
	--gradient_accumulation_steps="20" \
	--learning_rate="5e-5" \
	--warmup_steps="500" \
	--length_column_name="input_length" \
	--evaluation_strategy="steps" \
	--text_column_name="sentence" \
	--save_steps="500" \
	--eval_steps="500" \
	--logging_steps="50" \
	--layerdrop="0.0" \
	--activation_dropout="0.1" \
	--eval_metrics wer cer \
	--save_total_limit="3" \
	--feat_proj_dropout="0.0" \
	--mask_time_prob="0.75" \
	--mask_time_length="10" \
	--mask_feature_prob="0.25" \
	--mask_feature_length="64" \
	--freeze_feature_encoder \
	--chars_to_ignore « » ы — – ՚ ’ … , ? . ! - \; \: \" “ % ‘ \` ” � \
	--fp16 \
	--group_by_length \
	--use_auth_token \
	--report_to="tensorboard" \
	--do_train --do_eval \
	--gradient_checkpointing