#!/usr/bin/env bash
python run_flax_speech_recognition_seq2seq.py \
	--dataset_name="hf-internal-testing/librispeech_asr_dummy" \
	--model_name_or_path="./" \
	--dataset_config_name="clean" \
	--train_split_name="validation" \
	--eval_split_name="validation" \
	--test_split_name="validation[:90%]+validation[:95%]" \
	--output_dir="./" \
	--dataset_cache_dir="/home/sanchitgandhi/cache/huggingface/datasets" \
	--preprocessing_num_workers="1" \
	--length_column_name="input_length" \
	--overwrite_output_dir \
	--max_steps="10" \
	--eval_steps="5" \
	--save_steps="5" \
	--per_device_train_batch_size="2" \
	--per_device_eval_batch_size="2" \
	--logging_steps="1" \
	--max_duration_in_seconds="15" \
	--max_target_length="64" \
	--generation_max_length="40" \
	--generation_num_beams="1" \
	--final_generation_max_length="40" \
	--final_generation_num_beams="1" \
	--learning_rate="3e-4" \
	--warmup_steps="500" \
	--text_column_name="text" \
	--save_total_limit="1" \
        --wandb_project="flax-wav2vec2-2-bart-dummy" \
	--freeze_feature_encoder \
	--predict_with_generate \
	--do_lower_case \
       	--do_train \
	--do_eval \
	--do_predict \
	--push_to_hub