deepspeed run_speech_recognition_seq2seq_streaming.py \ --deepspeed="ds_config.json" \ --model_name_or_path="openai/whisper-medium" \ --datasets_config_file="/home/ubuntu/whisper-medium-is/ds_config.json" \ --language="icelandic" \ --train_split_name="train" \ --eval_dataset_name="language-and-voice-lab/samromur_asr" \ --eval_split_name="validation" \ --eval_steps="1000" \ --max_eval_samples="1000" \ --model_index_name="Whisper medium Icelandic" \ --max_steps="20000" \ --output_dir="./" \ --per_device_train_batch_size="64" \ --per_device_eval_batch_size="32" \ --logging_steps="25" \ --learning_rate="1e-5" \ --warmup_steps="500" \ --evaluation_strategy="steps" \ --save_strategy="steps" \ --save_steps="12000" \ --generation_max_length="225" \ --length_column_name="input_length" \ --max_duration_in_seconds="30" \ --text_column_name="normalized_text" \ --freeze_feature_encoder="False" \ --report_to="tensorboard" \ --metric_for_best_model="wer" \ --greater_is_better="False" \ --load_best_model_at_end \ --gradient_checkpointing \ --fp16 \ --do_train \ --do_eval \ --predict_with_generate \ --do_normalize_eval \ --streaming \ --push_to_hub \ --use_auth_token \ --mapper="/home/ubuntu/whisper/text_mapping"