File size: 1,382 Bytes
7acedc5 |
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 |
python run_speech_recognition_ctc.py \
--dataset_name="NbAiLab/NPSC" \
--model_name_or_path="facebook/wav2vec2-xls-r-300m" \
--dataset_config_name="16K_mp3" \
--data_cache_dir="/mnt/lv_ai_1_ficino/rolvb/cache" \
--output_dir="./" \
--overwrite_output_dir \
--num_train_epochs="20" \
--per_device_train_batch_size="16" \
--per_device_eval_batch_size="16" \
--gradient_accumulation_steps="4" \
--learning_rate="7.5e-5" \
--warmup_steps="1000" \
--length_column_name="input_length" \
--evaluation_strategy="steps" \
--text_column_name="text" \
--save_steps="250" \
--eval_steps="250" \
--logging_steps="50" \
--layerdrop="0.1" \
--activation_dropout="0.1" \
--save_total_limit="3" \
--freeze_feature_encoder \
--feat_proj_dropout="0.0" \
--mask_time_prob="0.75" \
--mask_time_length="10" \
--mask_feature_prob="0.25" \
--mask_feature_length="64" \
--chars_to_ignore , ? . ! \- \; \: \" “ % ‘ ” � — ’ … – _ \\ + \# / \
--gradient_checkpointing \
--use_auth_token \
--fp16 \
--group_by_length \
--do_train --do_eval \
--push_to_hub \
--preprocessing_num_workers="8" \
--run_name="baseline"
|