File size: 938 Bytes
1a10833 |
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 |
python run_speech_recognition_ctc.py \
--dataset_name="mozilla-foundation/common_voice_8_0" \
--dataset_config_name="nl" \
--text_column_name="sentence" \
--length_column_name="input_length" \
--use_auth_token \
--model_name_or_path="/workspace/hug_data/xls-r-2b-22-to-16.encoder.ctc48" \
--tokenizer_name_or_path="./" \
--output_dir="./output.step1" \
--overwrite_output_dir \
--num_train_epochs="5" \
--per_device_train_batch_size="8" \
--gradient_accumulation_steps="4" \
--learning_rate="5e-5" \
--warmup_steps="2000" \
--evaluation_strategy="steps" \
--save_steps="500" \
--eval_steps="500" \
--save_total_limit="4" \
--layerdrop="0.0" \
--freeze_feature_encoder \
--gradient_checkpointing \
--chars_to_ignore , ? . ! - \; \: \" “ % ‘ ” � _ \
--fp16 \
--group_by_length \
--do_train --do_eval
# --overwrite_output_dir \
|