Hugging Face's logo

abdusah
/
aradia-ctc-hubert-ft

Automatic Speech Recognition
Transformers
PyTorch
hubert
abdusahmbzuai/arabic_speech_massive_300hrs
Generated from Trainer
Inference Endpoints
Model card Files Community
Abdulwahab Sahyoun commited on
Commit
fe84a79
•
1 Parent(s): 505ed6e

update model card README.md

Browse files
Files changed (1) hide show
  1. README.md +94 -0
README.md ADDED
@@ -0,0 +1,94 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ tags:
4
+ - generated_from_trainer
5
+ model-index:
6
+ - name: aradia-ctc-hubert-ft
7
+ results: []
8
+ ---
9
+
10
+ <!-- This model card has been generated automatically according to the information the Trainer had access to. You
11
+ should probably proofread and complete it, then remove this comment. -->
12
+
13
+ # aradia-ctc-hubert-ft
14
+
15
+ This model is a fine-tuned version of [facebook/hubert-large-ll60k](https://huggingface.co/facebook/hubert-large-ll60k) on an unknown dataset.
16
+ It achieves the following results on the evaluation set:
17
+ - Loss: 0.6928
18
+ - Wer: 0.3946
19
+
20
+ ## Model description
21
+
22
+ More information needed
23
+
24
+ ## Intended uses & limitations
25
+
26
+ More information needed
27
+
28
+ ## Training and evaluation data
29
+
30
+ More information needed
31
+
32
+ ## Training procedure
33
+
34
+ ### Training hyperparameters
35
+
36
+ The following hyperparameters were used during training:
37
+ - learning_rate: 0.0003
38
+ - train_batch_size: 32
39
+ - eval_batch_size: 32
40
+ - seed: 42
41
+ - gradient_accumulation_steps: 2
42
+ - total_train_batch_size: 64
43
+ - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
44
+ - lr_scheduler_type: linear
45
+ - lr_scheduler_warmup_steps: 500
46
+ - num_epochs: 15.0
47
+ - mixed_precision_training: Native AMP
48
+
49
+ ### Training results
50
+
51
+ | Training Loss | Epoch | Step | Validation Loss | Wer |
52
+ |:-------------:|:-----:|:----:|:---------------:|:------:|
53
+ | No log | 0.43 | 100 | 3.6934 | 1.0 |
54
+ | No log | 0.87 | 200 | 3.0763 | 1.0 |
55
+ | No log | 1.3 | 300 | 2.9737 | 1.0 |
56
+ | No log | 1.74 | 400 | 2.5734 | 1.0 |
57
+ | 5.0957 | 2.17 | 500 | 1.1900 | 0.9011 |
58
+ | 5.0957 | 2.61 | 600 | 0.9726 | 0.7572 |
59
+ | 5.0957 | 3.04 | 700 | 0.8960 | 0.6209 |
60
+ | 5.0957 | 3.48 | 800 | 0.7851 | 0.5515 |
61
+ | 5.0957 | 3.91 | 900 | 0.7271 | 0.5115 |
62
+ | 1.0312 | 4.35 | 1000 | 0.7053 | 0.4955 |
63
+ | 1.0312 | 4.78 | 1100 | 0.6823 | 0.4737 |
64
+ | 1.0312 | 5.22 | 1200 | 0.6768 | 0.4595 |
65
+ | 1.0312 | 5.65 | 1300 | 0.6635 | 0.4488 |
66
+ | 1.0312 | 6.09 | 1400 | 0.6602 | 0.4390 |
67
+ | 0.6815 | 6.52 | 1500 | 0.6464 | 0.4310 |
68
+ | 0.6815 | 6.95 | 1600 | 0.6455 | 0.4394 |
69
+ | 0.6815 | 7.39 | 1700 | 0.6630 | 0.4312 |
70
+ | 0.6815 | 7.82 | 1800 | 0.6521 | 0.4126 |
71
+ | 0.6815 | 8.26 | 1900 | 0.6282 | 0.4284 |
72
+ | 0.544 | 8.69 | 2000 | 0.6248 | 0.4178 |
73
+ | 0.544 | 9.13 | 2100 | 0.6510 | 0.4104 |
74
+ | 0.544 | 9.56 | 2200 | 0.6527 | 0.4013 |
75
+ | 0.544 | 10.0 | 2300 | 0.6511 | 0.4064 |
76
+ | 0.544 | 10.43 | 2400 | 0.6734 | 0.4061 |
77
+ | 0.4478 | 10.87 | 2500 | 0.6756 | 0.4145 |
78
+ | 0.4478 | 11.3 | 2600 | 0.6727 | 0.3990 |
79
+ | 0.4478 | 11.74 | 2700 | 0.6619 | 0.4007 |
80
+ | 0.4478 | 12.17 | 2800 | 0.6614 | 0.4019 |
81
+ | 0.4478 | 12.61 | 2900 | 0.6695 | 0.4004 |
82
+ | 0.3919 | 13.04 | 3000 | 0.6778 | 0.3966 |
83
+ | 0.3919 | 13.48 | 3100 | 0.6872 | 0.3971 |
84
+ | 0.3919 | 13.91 | 3200 | 0.6882 | 0.3945 |
85
+ | 0.3919 | 14.35 | 3300 | 0.6938 | 0.3937 |
86
+ | 0.3919 | 14.78 | 3400 | 0.6928 | 0.3946 |
87
+
88
+
89
+ ### Framework versions
90
+
91
+ - Transformers 4.18.0.dev0
92
+ - Pytorch 1.10.2+cu113
93
+ - Datasets 1.18.4
94
+ - Tokenizers 0.11.6