File size: 1,107 Bytes
0f51336
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
#!/usr/bin/env bash
CUDA_VISIBLE_DEVICES=0 python run_speech_recognition_seq2seq_streaming.py \
	--model_name_or_path="openai/whisper-small" \
	--dataset_name="mozilla-foundation/common_voice_11_0" \
	--dataset_config_name="ru" \
	--language="Russian" \
	--train_split_name="train+validation" \
	--eval_split_name="test" \
	--model_index_name="Whisper Small Russian" \
	--max_steps="1000" \
	--output_dir="./" \
	--per_device_train_batch_size="32" \
	--per_device_eval_batch_size="16" \
	--logging_steps="25" \
	--learning_rate="1e-5" \
	--warmup_steps="50" \
	--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" \
	--lr_scheduler_type="constant_with_warmup" \
	--gradient_checkpointing \
	--fp16 \
	--overwrite_output_dir \
	--do_train \
	--do_eval \
	--predict_with_generate \
	--do_normalize_eval \
	--streaming \
	--use_auth_token \
	--push_to_hub