wav2vec2-xls-r-300m-japanese / run_training.sh
AndrewMcDowell's picture
Training in progress, step 1000
6bd1c39
raw
history blame
929 Bytes
python run_speech_recognition_ctc_bnb.py \
--dataset_name="mozilla-foundation/common_voice_8_0" \
--model_name_or_path="facebook/wav2vec2-xls-r-300m" \
--dataset_config_name="ja" \
--output_dir="./" \
--overwrite_output_dir \
--num_train_epochs="50" \
--per_device_train_batch_size="48" \
--per_device_eval_batch_size="8" \
--learning_rate="7.5e-5" \
--warmup_steps="2000" \
--length_column_name="input_length" \
--evaluation_strategy="steps" \
--text_column_name="sentence" \
--save_steps="1000" \
--eval_steps="1000" \
--logging_steps="100" \
--layerdrop="0.0" \
--activation_dropout="0.1" \
--save_total_limit="4" \
--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