infinitejoy
commited on
Commit
β’
c532f98
1
Parent(s):
99785a9
add training information
Browse files
README.md
CHANGED
@@ -38,7 +38,56 @@ More information needed
|
|
38 |
|
39 |
## Training and evaluation data
|
40 |
|
41 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
42 |
|
43 |
## Training procedure
|
44 |
|
|
|
38 |
|
39 |
## Training and evaluation data
|
40 |
|
41 |
+
Training machine details
|
42 |
+
|
43 |
+
- Platform: Linux-5.11.0-37-generic-x86_64-with-glibc2.10
|
44 |
+
- CPU cores: 60
|
45 |
+
- Python version: 3.8.8
|
46 |
+
- PyTorch version: 1.10.1+cu102
|
47 |
+
- GPU is visible: True
|
48 |
+
- Transformers version: 4.16.0.dev0
|
49 |
+
- Datasets version: 1.17.1.dev0
|
50 |
+
- soundfile version: 0.10.3
|
51 |
+
|
52 |
+
Training script
|
53 |
+
|
54 |
+
```bash
|
55 |
+
python run_speech_recognition_ctc.py \
|
56 |
+
--dataset_name="mozilla-foundation/common_voice_7_0" \
|
57 |
+
--model_name_or_path="facebook/wav2vec2-xls-r-300m" \
|
58 |
+
--dataset_config_name="or" \
|
59 |
+
--output_dir="./wav2vec2-large-xls-r-300m-odia" \
|
60 |
+
--overwrite_output_dir \
|
61 |
+
--num_train_epochs="120" \
|
62 |
+
--per_device_train_batch_size="16" \
|
63 |
+
--per_device_eval_batch_size="16" \
|
64 |
+
--gradient_accumulation_steps="2" \
|
65 |
+
--learning_rate="7.5e-5" \
|
66 |
+
--warmup_steps="500" \
|
67 |
+
--length_column_name="input_length" \
|
68 |
+
--evaluation_strategy="steps" \
|
69 |
+
--text_column_name="sentence" \
|
70 |
+
--chars_to_ignore , ? . ! \- \; \: \" β % β β οΏ½ β \β β¦ \β \' \β \β \
|
71 |
+
--save_steps="500" \
|
72 |
+
--eval_steps="500" \
|
73 |
+
--logging_steps="100" \
|
74 |
+
--layerdrop="0.0" \
|
75 |
+
--activation_dropout="0.1" \
|
76 |
+
--save_total_limit="3" \
|
77 |
+
--freeze_feature_encoder \
|
78 |
+
--feat_proj_dropout="0.0" \
|
79 |
+
--mask_time_prob="0.75" \
|
80 |
+
--mask_time_length="10" \
|
81 |
+
--mask_feature_prob="0.25" \
|
82 |
+
--mask_feature_length="64" \
|
83 |
+
--gradient_checkpointing \
|
84 |
+
--use_auth_token \
|
85 |
+
--fp16 \
|
86 |
+
--group_by_length \
|
87 |
+
--do_train --do_eval \
|
88 |
+
--push_to_hub
|
89 |
+
```
|
90 |
+
|
91 |
|
92 |
## Training procedure
|
93 |
|