cmd.sh

torchrun \
	--nproc_per_node 2 run_speech_recognition_ctc_adapter.py \
	--model_name_or_path="facebook/mms-1b-all" \
	--dataset_name="/mnt/md0/synvoices/data/naijavoices_500h" \
	--train_split_name="train" \
	--eval_split_name="validation" \
	--output_dir="./training_runs/mms-1b-naijavoices_500h-hau-ft" \
	--num_train_epochs="2" \
	--per_device_train_batch_size="8" \
	--per_device_eval_batch_size="8" \
    --gradient_accumulation_steps="2" \
	--learning_rate="1e-3" \
	--warmup_steps="100" \
	--eval_strategy="steps" \
	--save_steps="500" \
	--eval_steps="500" \
	--logging_steps="1" \
	--eval_metrics wer cer \
	--save_total_limit="2" \
    --max_duration_in_seconds="30" \
	--target_language="hau" \
	--overwrite_lang_vocab \
	--gradient_checkpointing \
	--fp16 \
	--do_train --do_eval \
	--preprocessing_num_workers="12" \
	--dataloader_num_workers="12"