File size: 1,037 Bytes
630b73d 1ad5896 f4949f5 630b73d f4949f5 630b73d 4694f52 630b73d 4694f52 |
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 |
python run_speech_recognition_ctc_bnb.py \
--model_name_or_path="facebook/wav2vec2-xls-r-2b" \
--dataset_name="openslr_SLR66" \
--train_split_name="train" \
--preprocessing_num_workers="8" \
--output_dir="./" \
--overwrite_output_dir \
--num_train_epochs="150" \
--per_device_train_batch_size="4" \
--per_device_eval_batch_size="4" \
--gradient_accumulation_steps="12" \
--learning_rate="3e-6" \
--warmup_steps="2000" \
--length_column_name="input_length" \
--evaluation_strategy="steps" \
--text_column_name="sentence" \
--chars_to_ignore , ? . ! \- \; \: \" “ % ‘ ” � — ’ … – \
--save_steps="500" \
--eval_steps="500" \
--logging_steps="10" \
--layerdrop="0.0" \
--hidden_dropout="0.15" \
--activation_dropout="0.1" \
--save_total_limit="2" \
--freeze_feature_encoder \
--feat_proj_dropout="0.0" \
--mask_time_prob="0.4" \
--mask_time_length="10" \
--mask_feature_prob="0.25" \
--mask_feature_length="64" \
--gradient_checkpointing \
--use_auth_token \
--fp16 \
--group_by_length \
--do_train --do_eval \
--push_to_hub
|