| python $1run_speech_recognition_seq2seq_streaming.py \ | |
| --model_name_or_path="openai/whisper-medium" \ | |
| --dataset_train_name="mozilla-foundation/common_voice_11_0,mozilla-foundation/common_voice_11_0,mozilla-foundation/common_voice_11_0,babelbox/babelbox_voice,NbAiLab/NST,NbAiLab/NPSC,google/fleurs,google/fleurs,google/fleurs" \ | |
| --dataset_train_config_name="sv-SE,da,nn-NO,nst,no-distant,16K_mp3_nynorsk,sv_se,da_dk,nb_no" \ | |
| --language_train="sv,da,no,sv,no,no,sv,da,no" \ | |
| --train_split_name="train+validation,train+validation,train+validation,train,train+test,train+validation,train+validation,train+validation,train+validation" \ | |
| --dataset_eval_name="mozilla-foundation/common_voice_11_0" \ | |
| --dataset_eval_config_name="sv-SE" \ | |
| --language_eval="sv" \ | |
| --eval_split_name="test" \ | |
| --model_index_name="Whisper Medium Nordic" \ | |
| --num_train_epochs="1" \ | |
| --output_dir="./" \ | |
| --per_device_train_batch_size="32" \ | |
| --per_device_eval_batch_size="16" \ | |
| --logging_steps="25" \ | |
| --learning_rate="3e-6" \ | |
| --warmup_ratio="0.1" \ | |
| --evaluation_strategy="steps" \ | |
| --eval_steps="1000" \ | |
| --save_strategy="steps" \ | |
| --save_steps="1000" \ | |
| --generation_max_length="225" \ | |
| --length_column_name="input_length" \ | |
| --max_duration_in_seconds="30" \ | |
| --text_column_name="sentence,text,raw_transcription" \ | |
| --freeze_feature_encoder="False" \ | |
| --report_to="wandb" \ | |
| --save_total_limit="2" \ | |
| --metric_for_best_model="wer" \ | |
| --greater_is_better="False" \ | |
| --load_best_model_at_end \ | |
| --gradient_checkpointing \ | |
| --overwrite_output_dir \ | |
| --do_train \ | |
| --do_eval \ | |
| --fp16 \ | |
| --predict_with_generate \ | |
| --do_normalize_eval \ | |
| --streaming \ | |
| --use_auth_token |