File size: 1,257 Bytes
8f1a9b5
 
63865a1
 
8f1a9b5
63865a1
 
 
8f1a9b5
 
63865a1
8f1a9b5
 
 
 
 
 
 
 
 
 
 
 
 
63865a1
8f1a9b5
63865a1
8f1a9b5
 
 
 
 
 
 
 
63865a1
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
python run_speech_recognition_seq2seq_streaming.py \
	--model_name_or_path="marinone94/whisper-medium-nordic" \
	--dataset_train_name="mozilla-foundation/common_voice_11_0,babelbox/babelbox_voice,google/fleurs" \
	--dataset_train_config_name="sv-SE,nst,sv_se" \
	--language="swedish" \
	--train_split_name="train+validation,train,train+validation+test" \
	--dataset_eval_name="mozilla-foundation/common_voice_11_0" \
	--dataset_eval_config_name="sv-SE" \
	--eval_split_name="test" \
	--model_index_name="Whisper Medium Swedish" \
	--max_steps="5000" \
	--output_dir="./" \
	--per_device_train_batch_size="32" \
	--per_device_eval_batch_size="16" \
	--logging_steps="25" \
	--learning_rate="1e-5" \
	--warmup_steps="250" \
	--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,raw_transcription" \
	--freeze_feature_encoder="False" \
	--report_to="wandb" \
	--metric_for_best_model="wer" \
	--greater_is_better="False" \
	--load_best_model_at_end \
	--gradient_checkpointing \
	--fp16 \
	--predict_with_generate \
	--do_normalize_eval \
	--streaming \
	--use_auth_token