35 lines
1.0 KiB
Bash
35 lines
1.0 KiB
Bash
python run_speech_recognition_ctc.py \
|
||
--dataset_name="mozilla-foundation/common_voice_7_0" \
|
||
--model_name_or_path="facebook/wav2vec2-xls-r-300m" \
|
||
--dataset_config_name="cy" \
|
||
--output_dir="./wav2vec2-large-xls-r-300m-welsh" \
|
||
--overwrite_output_dir \
|
||
--num_train_epochs="50" \
|
||
--per_device_train_batch_size="32" \
|
||
--per_device_eval_batch_size="16" \
|
||
--gradient_accumulation_steps="1" \
|
||
--learning_rate="7e-5" \
|
||
--warmup_steps="3000" \
|
||
--length_column_name="input_length" \
|
||
--evaluation_strategy="steps" \
|
||
--text_column_name="sentence" \
|
||
--chars_to_ignore , ? . ! \- \; \: \" “ % ‘ ” <20> — ’ … – \' \
|
||
--save_steps="3000" \
|
||
--eval_steps="3000" \
|
||
--logging_steps="100" \
|
||
--layerdrop="0.0" \
|
||
--activation_dropout="0.1" \
|
||
--save_total_limit="2" \
|
||
--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" \
|
||
--gradient_checkpointing \
|
||
--use_auth_token \
|
||
--fp16 \
|
||
--group_by_length \
|
||
--do_train --do_eval \
|
||
--push_to_hub
|