marinone94 commited on
Commit
868234c
1 Parent(s): d73da22

script to continue training

Browse files
Files changed (2) hide show
  1. README_TEMPLATE.md +6 -1
  2. cont-run.sh +35 -0
README_TEMPLATE.md CHANGED
@@ -47,4 +47,9 @@ model-index:
47
 
48
  This model is a fine-tuned version of [KBLab/wav2vec2-large-voxrex](https://huggingface.co/KBLab/wav2vec2-large-voxrex) on 2 epochs of the MARINONE94/NST_SV - SV dataset (80% random split with seed 42 as the dataset for now has only the "train" split), and then on 50 epochs of the the MOZILLA-FOUNDATION/COMMON_VOICE_8_0 - SV-SE dataset ("train+validation" split).
49
  See run.sh to have a complete overview of all the training steps.
50
- NOTE: the first training for now didn't work as expected, so it might be useless or even degrade performance. Further investigation and development is needed.
 
 
 
 
 
 
47
 
48
  This model is a fine-tuned version of [KBLab/wav2vec2-large-voxrex](https://huggingface.co/KBLab/wav2vec2-large-voxrex) on 2 epochs of the MARINONE94/NST_SV - SV dataset (80% random split with seed 42 as the dataset for now has only the "train" split), and then on 50 epochs of the the MOZILLA-FOUNDATION/COMMON_VOICE_8_0 - SV-SE dataset ("train+validation" split).
49
  See run.sh to have a complete overview of all the training steps.
50
+ NOTE: the first training for now didn't work as expected, so it might be useless or even degrade performance. Further investigation and development is needed.
51
+
52
+ d73da225cfdc57213ea4ab67b24bb87ac41f4392 is the commit at the end of the first training:
53
+ ```
54
+ sh run.sh
55
+ ```
cont-run.sh ADDED
@@ -0,0 +1,35 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ python run_speech_recognition_ctc.py \
2
+ --dataset_name="mozilla-foundation/common_voice_9_0" \
3
+ --model_name_or_path="KBLab/wav2vec2-large-voxrex" \
4
+ --dataset_config_name="sv-SE" \
5
+ --train_split_name="train+validation" \
6
+ --eval_split_name="test" \
7
+ --output_dir="./" \
8
+ --num_train_epochs="150" \
9
+ --per_device_train_batch_size="32" \
10
+ --per_device_eval_batch_size="32" \
11
+ --gradient_accumulation_steps="4" \
12
+ --learning_rate="7.5e-4" \
13
+ --length_column_name="input_length" \
14
+ --evaluation_strategy="steps" \
15
+ --save_strategy="steps" \
16
+ --eval_steps="1000" \
17
+ --save_steps="1000" \
18
+ --text_column_name="sentence" \
19
+ --chars_to_ignore , ? . ! \- \; \: \" “ % ‘ ” � — ’ … – \
20
+ --logging_steps="100" \
21
+ --layerdrop="0.0" \
22
+ --activation_dropout="0.15" \
23
+ --save_total_limit="2" \
24
+ --freeze_feature_encoder \
25
+ --feat_proj_dropout="0.0" \
26
+ --mask_time_prob="0.75" \
27
+ --mask_time_length="10" \
28
+ --mask_feature_prob="0.25" \
29
+ --mask_feature_length="64" \
30
+ --gradient_checkpointing \
31
+ --use_auth_token \
32
+ --fp16 \
33
+ --group_by_length \
34
+ --do_train --do_eval \
35
+ --push_to_hub