samitizerxu commited on
Commit
fe18188
1 Parent(s): ace4f16

Added training run command

Browse files
Files changed (1) hide show
  1. run.sh +36 -0
run.sh ADDED
@@ -0,0 +1,36 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/bin/bash
2
+
3
+ python run_speech_recognition_ctc.py \
4
+ --dataset_name="common_voice" \
5
+ --model_name_or_path="facebook/wav2vec2-xls-r-300m" \
6
+ --dataset_config_name="zh-CN" \
7
+ --output_dir="./wav2vec2-xls-r-300m-zh-CN" \
8
+ --overwrite_output_dir \
9
+ --num_train_epochs="50" \
10
+ --per_device_train_batch_size="8" \
11
+ --per_device_eval_batch_size="8" \
12
+ --max_duration_in_seconds="7" \
13
+ --max_eval_samples="3000" \
14
+ --gradient_accumulation_steps="4" \
15
+ --learning_rate="7.5e-5" \
16
+ --warmup_steps="2000" \
17
+ --length_column_name="input_length" \
18
+ --evaluation_strategy="steps" \
19
+ --text_column_name="sentence" \
20
+ --save_steps="500" \
21
+ --eval_steps="500" \
22
+ --logging_steps="100" \
23
+ --layerdrop="0.0" \
24
+ --activation_dropout="0.1" \
25
+ --save_total_limit="3" \
26
+ --freeze_feature_encoder \
27
+ --feat_proj_dropout="0.0" \
28
+ --mask_time_prob="0.75" \
29
+ --mask_time_length="10" \
30
+ --mask_feature_prob="0.25" \
31
+ --mask_feature_length="64" \
32
+ --gradient_checkpointing \
33
+ --fp16 \
34
+ --group_by_length \
35
+ --do_train --do_eval \
36
+