File size: 1,222 Bytes
9fcc5b4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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 \
    --output_dir="./" \
    --model_name_or_path="openai/whisper-small" \
    --model_index_name="Whisper Small Japanese" \
    --dataset_name="mozilla-foundation/common_voice_11_0" \
    --dataset_config_name="ja" \
    --language="japanese" \
    --train_split_name="train+validation" \
    --eval_split_name="test" \
    --learning_rate="1e-5" \
    --per_device_train_batch_size="64" \
    --per_device_eval_batch_size="32" \
    --max_steps="1000" \
    --warmup_steps="500" \
    --logging_steps="25" \
    --evaluation_strategy="steps" \
    --eval_steps="200" \
    --save_strategy="steps" \
    --save_steps="200" \
    --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" \
    --load_best_model_at_end \
    --gradient_checkpointing \
    --fp16 \
    --overwrite_output_dir \
    --do_train \
    --do_eval \
    --predict_with_generate \
    --do_normalize_eval \
    --streaming \
    --use_auth_token \
    --push_to_hub