Hugging Face's logo

gopikachu
/
whisper-small-hi

Automatic Speech Recognition
Transformers
TensorBoard
Safetensors
Hindi
whisper
Generated from Trainer
Eval Results
Inference Endpoints
Model card Files Metrics Community
gopikachu commited on
Commit
7664cc2
•
1 Parent(s): d75a273

End of training

Browse files
Files changed (3) hide show
  1. README.md +23 -23
  2. generation_config.json +22 -17
  3. model.safetensors +1 -1
README.md CHANGED
@@ -2,7 +2,7 @@
2
  language:
3
  - hi
4
  license: apache-2.0
5
- base_model: openai/whisper-large
6
  tags:
7
  - generated_from_trainer
8
  datasets:
@@ -10,7 +10,7 @@ datasets:
10
  metrics:
11
  - wer
12
  model-index:
13
- - name: Whisper Large Hi - Gopika
14
  results:
15
  - task:
16
  name: Automatic Speech Recognition
@@ -22,18 +22,18 @@ model-index:
22
  metrics:
23
  - name: Wer
24
  type: wer
25
- value: 96.73659673659674
26
  ---
27
 
28
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
29
  should probably proofread and complete it, then remove this comment. -->
30
 
31
- # Whisper Large Hi - Gopika
32
 
33
- This model is a fine-tuned version of [openai/whisper-large](https://huggingface.co/openai/whisper-large) on the Konnakol dataset.
34
  It achieves the following results on the evaluation set:
35
- - Loss: 2.2174
36
- - Wer: 96.7366
37
 
38
  ## Model description
39
 
@@ -52,30 +52,30 @@ More information needed
52
  ### Training hyperparameters
53
 
54
  The following hyperparameters were used during training:
55
- - learning_rate: 1e-05
56
- - train_batch_size: 16
57
- - eval_batch_size: 8
58
  - seed: 42
59
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
60
  - lr_scheduler_type: linear
61
  - lr_scheduler_warmup_steps: 500
62
- - training_steps: 10000
63
  - mixed_precision_training: Native AMP
64
 
65
  ### Training results
66
 
67
- | Training Loss | Epoch | Step | Validation Loss | Wer |
68
- |:-------------:|:--------:|:-----:|:---------------:|:-------:|
69
- | 0.0735 | 21.7391 | 1000 | 1.3101 | 93.7063 |
70
- | 0.0146 | 43.4783 | 2000 | 1.7063 | 87.5291 |
71
- | 0.0059 | 65.2174 | 3000 | 1.9682 | 94.1725 |
72
- | 0.0072 | 86.9565 | 4000 | 1.9777 | 93.0070 |
73
- | 0.0036 | 108.6957 | 5000 | 2.0244 | 96.9697 |
74
- | 0.0021 | 130.4348 | 6000 | 2.0688 | 96.9697 |
75
- | 0.0025 | 152.1739 | 7000 | 2.1008 | 96.3869 |
76
- | 0.0028 | 173.9130 | 8000 | 2.1506 | 97.0862 |
77
- | 0.0034 | 195.6522 | 9000 | 2.1841 | 97.9021 |
78
- | 0.0026 | 217.3913 | 10000 | 2.2174 | 96.7366 |
79
 
80
 
81
  ### Framework versions
 
2
  language:
3
  - hi
4
  license: apache-2.0
5
+ base_model: openai/whisper-small
6
  tags:
7
  - generated_from_trainer
8
  datasets:
 
10
  metrics:
11
  - wer
12
  model-index:
13
+ - name: Whisper Small Hi - Gopika Krishnan
14
  results:
15
  - task:
16
  name: Automatic Speech Recognition
 
22
  metrics:
23
  - name: Wer
24
  type: wer
25
+ value: 87.64568764568764
26
  ---
27
 
28
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
29
  should probably proofread and complete it, then remove this comment. -->
30
 
31
+ # Whisper Small Hi - Gopika Krishnan
32
 
33
+ This model is a fine-tuned version of [openai/whisper-small](https://huggingface.co/openai/whisper-small) on the Konnakol dataset.
34
  It achieves the following results on the evaluation set:
35
+ - Loss: 2.2352
36
+ - Wer: 87.6457
37
 
38
  ## Model description
39
 
 
52
  ### Training hyperparameters
53
 
54
  The following hyperparameters were used during training:
55
+ - learning_rate: 0.0001
56
+ - train_batch_size: 32
57
+ - eval_batch_size: 16
58
  - seed: 42
59
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
60
  - lr_scheduler_type: linear
61
  - lr_scheduler_warmup_steps: 500
62
+ - training_steps: 5000
63
  - mixed_precision_training: Native AMP
64
 
65
  ### Training results
66
 
67
+ | Training Loss | Epoch | Step | Validation Loss | Wer |
68
+ |:-------------:|:--------:|:----:|:---------------:|:-------:|
69
+ | 0.1507 | 21.7391 | 500 | 1.2891 | 87.7622 |
70
+ | 0.0428 | 43.4783 | 1000 | 1.4133 | 93.7063 |
71
+ | 0.0111 | 65.2174 | 1500 | 1.7252 | 89.3939 |
72
+ | 0.0063 | 86.9565 | 2000 | 1.8134 | 85.8974 |
73
+ | 0.0035 | 108.6957 | 2500 | 2.0195 | 85.7809 |
74
+ | 0.003 | 130.4348 | 3000 | 2.0771 | 87.8788 |
75
+ | 0.0027 | 152.1739 | 3500 | 2.1378 | 87.5291 |
76
+ | 0.0025 | 173.9130 | 4000 | 2.1730 | 86.4802 |
77
+ | 0.0025 | 195.6522 | 4500 | 2.2126 | 87.8788 |
78
+ | 0.0025 | 217.3913 | 5000 | 2.2352 | 87.6457 |
79
 
80
 
81
  ### Framework versions
generation_config.json CHANGED
@@ -1,40 +1,44 @@
1
  {
2
  "alignment_heads": [
3
  [
4
- 9,
5
- 19
6
  ],
7
  [
8
- 11,
9
- 2
10
  ],
11
  [
12
- 11,
13
- 4
14
  ],
15
  [
16
- 11,
17
- 17
18
  ],
19
  [
20
- 22,
21
  7
22
  ],
23
  [
24
- 22,
25
- 11
26
  ],
27
  [
28
- 22,
29
- 17
30
  ],
31
  [
32
- 23,
33
- 2
 
 
 
 
34
  ],
35
  [
36
- 23,
37
- 15
38
  ]
39
  ],
40
  "begin_suppress_tokens": [
@@ -146,6 +150,7 @@
146
  "<|yo|>": 50325,
147
  "<|zh|>": 50260
148
  },
 
149
  "max_initial_timestamp_index": 50,
150
  "max_length": 448,
151
  "no_timestamps_token_id": 50363,
 
1
  {
2
  "alignment_heads": [
3
  [
4
+ 5,
5
+ 3
6
  ],
7
  [
8
+ 5,
9
+ 9
10
  ],
11
  [
12
+ 8,
13
+ 0
14
  ],
15
  [
16
+ 8,
17
+ 4
18
  ],
19
  [
20
+ 8,
21
  7
22
  ],
23
  [
24
+ 8,
25
+ 8
26
  ],
27
  [
28
+ 9,
29
+ 0
30
  ],
31
  [
32
+ 9,
33
+ 7
34
+ ],
35
+ [
36
+ 9,
37
+ 9
38
  ],
39
  [
40
+ 10,
41
+ 5
42
  ]
43
  ],
44
  "begin_suppress_tokens": [
 
150
  "<|yo|>": 50325,
151
  "<|zh|>": 50260
152
  },
153
+ "language": "hindi",
154
  "max_initial_timestamp_index": 50,
155
  "max_length": 448,
156
  "no_timestamps_token_id": 50363,
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:4e602d4f178354e6d3a11f49aff77724faa4466f5751f2a221487d031e13c3f7
3
  size 966995080
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:072032d04333e568c87a151ce6cab4f820531b1218076deecdc8775e0250163e
3
  size 966995080