File size: 1,413 Bytes
4ee7109 |
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 |
#!/usr/bin/env bash
CUDA_VISIBLE_DEVICES=1 python run_speech_recognition_rnnt.py \
--config_path="conf/conformer_transducer_bpe_xlarge.yaml" \
--model_name_or_path="stt_en_conformer_transducer_xlarge" \
--dataset_name="kensho/spgispeech" \
--tokenizer_path="tokenizer" \
--vocab_size="1024" \
--num_train_epochs="3" \
--dataset_config_name="S" \
--train_split_name="train" \
--eval_split_name="validation" \
--test_split_name="test" \
--text_column_name="transcript" \
--output_dir="./" \
--run_name="rnnt-spgispeech-1e-3-freeze-encoder-unfreeze-encoder" \
--wandb_project="rnnt-debug-spgispeech" \
--per_device_train_batch_size="8" \
--per_device_eval_batch_size="4" \
--logging_steps="25" \
--learning_rate="1e-3" \
--warmup_steps="500" \
--save_strategy="no" \
--evaluation_strategy="epoch" \
--report_to="wandb" \
--preprocessing_num_workers="4" \
--fused_batch_size="8" \
--length_column_name="input_lengths" \
--final_decoding_strategy="greedy_batch" \
--final_num_beams="1" \
--do_lower_case="False" \
--freeze_encoder \
--fuse_loss_wer \
--group_by_length \
--overwrite_output_dir \
--do_train \
--do_eval \
--use_auth_token
|