File size: 1,079 Bytes
1e5383d |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 |
python run_speech_recognition_seq2seq_streaming.py \
--model_name_or_path="openai/whisper-large-v3" \
--dataset_name="rishabhjain16/owr_cv_albanian" \
--dataset_config_name="default" \
--language="Albanian" \
--train_split_name="train" \
--eval_split_name="test" \
--model_index_name="Whisper large V3 to CV Albanian" \
--max_steps="5000" \
--output_dir="./" \
--per_device_train_batch_size="16" \
--per_device_eval_batch_size="16" \
--logging_steps="25" \
--learning_rate="1e-5" \
--warmup_steps="500" \
--evaluation_strategy="steps" \
--eval_steps="500" \
--save_strategy="steps" \
--save_steps="1000" \
--generation_max_length="225" \
--length_column_name="input_length" \
--max_duration_in_seconds="30" \
--text_column_name="transcription" \
--freeze_feature_encoder="False" \
--report_to="tensorboard" \
--metric_for_best_model="wer" \
--greater_is_better="False" \
--load_best_model_at_end \
--gradient_checkpointing \
--fp16 \
--overwrite_output_dir \
--do_train \
--do_eval \
--predict_with_generate \
--do_normalize_eval \
--streaming \
--use_auth_token \
--push_to_hub
|