File size: 1,991 Bytes
80a5c4d |
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 40 41 42 43 |
export LD_LIBRARY_PATH=/usr/lib/x86_64-linux-gnu
deepspeed run_speech_recognition_seq2seq_streaming.py \
--deepspeed="ds_config.json" \
--model_name_or_path="openai/whisper-large-v2" \
--train_dataset_names "mozilla-foundation/common_voice_11_0" "mozilla-foundation/common_voice_11_0" "/home/ubuntu/fleurs_stream.csv" "/home/ubuntu/css10_stream.csv" "Finnish-NLP/youtube_fi_110k_preprocessed" "Finnish-NLP/voxpopuli_fi_audio_processed" "Finnish-NLP/parliament_audio_processed" "Finnish-NLP/aalto_eduskunta_asr_audio_processed" "Finnish-NLP/fbc_dialog_audio_processed" "Finnish-NLP/fbc_monolog_audio_processed" \
--train_dataset_config_names "fi" "fi" "" "" "" "" "" "" "" "" \
--train_dataset_split_name "train" "validation" "train" "train" "train" "train" "train" "train" "train" "train" \
--train_dataset_text_column_names "sentence" "sentence" "raw_transcription" "sentence_corrected" "texts" "sentence" "sentence" "sentence" "sentence" "sentence" \
--eval_dataset_name="mozilla-foundation/common_voice_11_0" \
--eval_dataset_config_name="fi" \
--eval_split_name="test" \
--language="finnish" \
--model_index_name="Whisper Large V2 Finnish" \
--max_steps="30000" \
--output_dir="./" \
--per_device_train_batch_size="64" \
--per_device_eval_batch_size="16" \
--gradient_accumulation_steps="2" \
--logging_steps="25" \
--learning_rate="3e-6" \
--warmup_steps="500" \
--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" \
--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 \
--use_auth_token \
--push_to_hub |