infinitejoy commited on
Commit
2635b09
1 Parent(s): fafb9bc

training code

Browse files
Files changed (2) hide show
  1. run.sh +34 -0
  2. run_speech_recognition_ctc.py +0 -3
run.sh ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ python run_speech_recognition_ctc.py \
2
+ --dataset_name="mozilla-foundation/common_voice_8_0" \
3
+ --model_name_or_path="facebook/wav2vec2-xls-r-300m" \
4
+ --dataset_config_name="ab" \
5
+ --output_dir="./wav2vec2-large-xls-r-300m-abkhaz-cv8" \
6
+ --overwrite_output_dir \
7
+ --num_train_epochs="50" \
8
+ --per_device_train_batch_size="32" \
9
+ --per_device_eval_batch_size="16" \
10
+ --gradient_accumulation_steps="1" \
11
+ --learning_rate="7e-5" \
12
+ --warmup_steps="4000" \
13
+ --length_column_name="input_length" \
14
+ --evaluation_strategy="steps" \
15
+ --text_column_name="sentence" \
16
+ --chars_to_ignore , ? . ! \- \; \: \" “ % ‘ ” � — ’ … – \' \c \
17
+ --save_steps="4000" \
18
+ --eval_steps="4000" \
19
+ --logging_steps="1000" \
20
+ --layerdrop="0.0" \
21
+ --activation_dropout="0.1" \
22
+ --save_total_limit="2" \
23
+ --freeze_feature_encoder \
24
+ --feat_proj_dropout="0.0" \
25
+ --mask_time_prob="0.75" \
26
+ --mask_time_length="10" \
27
+ --mask_feature_prob="0.25" \
28
+ --mask_feature_length="64" \
29
+ --gradient_checkpointing \
30
+ --use_auth_token \
31
+ --fp16 \
32
+ --group_by_length \
33
+ --do_train --do_eval \
34
+ --push_to_hub
run_speech_recognition_ctc.py CHANGED
@@ -444,9 +444,6 @@ def main():
444
  batch["target_text"] = re.sub(chars_to_ignore_regex, "", batch[text_column_name]).lower() + " "
445
  else:
446
  batch["target_text"] = batch[text_column_name].lower() + " "
447
-
448
- batch["target_text"] = batch[text_column_name].lower().replace('c', '')
449
-
450
  return batch
451
 
452
  with training_args.main_process_first(desc="dataset map special characters removal"):
 
444
  batch["target_text"] = re.sub(chars_to_ignore_regex, "", batch[text_column_name]).lower() + " "
445
  else:
446
  batch["target_text"] = batch[text_column_name].lower() + " "
 
 
 
447
  return batch
448
 
449
  with training_args.main_process_first(desc="dataset map special characters removal"):