File size: 1,108 Bytes
b51985e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 run_speech_recognition_seq2seq_streaming.py \
--model_name_or_path="openai/whisper-large-v2" \
--dataset_name="mozilla-foundation/common_voice_11_0" \
--dataset_config_name="vi" \
--language="vietnamese" \
--train_split_name="train+validation" \
--eval_split_name="test" \
--model_index_name="Whisper Large V2 Vietnamese" \
--max_steps="10000" \
--output_dir="./" \
--per_device_train_batch_size="8" \
--per_device_eval_batch_size="4" \
--logging_steps="25" \
--learning_rate="1e-5" \
--warmup_steps="1000" \
--evaluation_strategy="steps" \
--eval_steps="1000" \
--save_strategy="steps" \
--save_steps="1000" \
--generation_max_length="225" \
--length_column_name="input_length" \
--max_duration_in_seconds="30" \
--text_column_name="sentence" \
--freeze_feature_encoder="False" \
--report_to="tensorboard" \
--metric_for_best_model="wer" \
--greater_is_better="False" \
--gradient_accumulation_steps="2" \
--load_best_model_at_end \
--gradient_checkpointing \
--fp16 \
--overwrite_output_dir \
--do_train \
--do_eval \
--predict_with_generate \
--do_normalize_eval \
--use_auth_token \
--push_to_hub