File size: 1,173 Bytes
98e5ae5 841ba8a 98e5ae5 841ba8a ec78bff 841ba8a 37570e5 98e5ae5 841ba8a ec78bff 98e5ae5 841ba8a 98e5ae5 ec78bff 841ba8a ec78bff 98e5ae5 841ba8a ec78bff 841ba8a |
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 ~/xls-r-uzbek-cv8/run_speech_recognition_ctc.py \
--dataset_name="mozilla-foundation/common_voice_8_0" \
--model_name_or_path="facebook/wav2vec2-xls-r-300m" \
--dataset_config_name="uz" \
--output_dir="~/xls-r-uzbek-cv8" \
--train_split_name="train[:30%]" \
--eval_split_name="validation[50%:]" \
--overwrite_output_dir \
--num_train_epochs="50" \
--per_device_train_batch_size="64" \
--per_device_eval_batch_size="8" \
--gradient_accumulation_steps="4" \
--learning_rate="3e-5" \
--warmup_steps="500" \
--length_column_name="input_length" \
--evaluation_strategy="steps" \
--text_column_name="sentence" \
--eval_metrics wer cer \
--save_steps="200" \
--eval_steps="200" \
--logging_steps="100" \
--min_duration_in_seconds="0.2" \
--layerdrop="0.05" \
--activation_dropout="0.1" \
--save_total_limit="3" \
--freeze_feature_encoder \
--feat_proj_dropout="0.05" \
--mask_time_prob="0.50" \
--mask_time_length="10" \
--mask_feature_prob="0.15" \
--mask_feature_length="64" \
--gradient_checkpointing \
--use_auth_token \
--fp16 \
--group_by_length \
--do_train --do_eval \
--push_to_hub
# --chars_to_ignore \ # default to all punct
|