File size: 1,206 Bytes
f3f904f
9ff1856
f3f904f
 
9ff1856
 
 
f3f904f
 
 
 
 
 
 
 
 
 
 
 
 
 
9ff1856
f3f904f
 
 
 
9ff1856
f3f904f
 
 
 
 
 
 
 
 
 
9ff1856
f3f904f
 
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
40
41
python run_speech_recognition_ctc.py \
	--dataset_name /workspace/data/uk/composed_dataset/ \
	--train_split_name train \
	--model_name_or_path="facebook/wav2vec2-xls-r-1b" \
	--output_dir ./ \
	--overwrite_output_dir \
	--max_steps 12000 \
	--per_device_train_batch_size="16" \
	--per_device_eval_batch_size="64" \
	--gradient_accumulation_steps="8" \
	--dataloader_num_workers 8 \
	--learning_rate="5e-5" \
	--adam_beta2 0.98 \
	--lr_scheduler_type cosine \
	--warmup_ratio 0.1 \
	--evaluation_strategy="steps" \
	--text_column_name="sentence" \
	--chars_to_ignore \, \? \. \! \- \; \: \" \“ \% \‘ \” \� \' « » \( \) ՝ ՛ ՚ \– \— \… ý \
	--save_steps="500" \
	--eval_steps="500" \
	--logging_steps="100" \
	--save_total_limit 10 \
	--freeze_feature_encoder \
	--layerdrop="0.1" \
	--activation_dropout="0.1" \
	--feat_proj_dropout="0.0" \
	--mask_time_prob="0.7" \
	--mask_time_length="10" \
	--mask_feature_prob="0.25" \
	--mask_feature_length="64" \
	--gradient_checkpointing \
	--use_auth_token \
	--fp16 \
	--group_by_length \
	--do_train --do_eval \
	--load_best_model_at_end \
	--report_to all \
	--run_name="xlsr-uk-1b-1" \
	--wandb_project="xlsr-uk" \
	--bnb --tristage_sched