Hugging Face's logo

Tomeno
/
whisper-tiny-tok-synth-1

Automatic Speech Recognition
Transformers
TensorBoard
Safetensors
Toki Pona
whisper
Generated from Trainer
Model card Files
xet
Metrics Community
Tomeno commited on
Commit
c869164
·
verified ·
1 Parent(s): 1fb9fce

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +69 -76
README.md CHANGED
@@ -1,76 +1,69 @@
1
- ---
2
- library_name: transformers
3
- language:
4
- - tok
5
- license: apache-2.0
6
- base_model: openai/whisper-tiny
7
- tags:
8
- - generated_from_trainer
9
- metrics:
10
- - wer
11
- model-index:
12
- - name: Whisper Tiny - Toki Pona - Synthetic Test 1
13
- results: []
14
- ---
15
-
16
- <!-- This model card has been generated automatically according to the information the Trainer had access to. You
17
- should probably proofread and complete it, then remove this comment. -->
18
-
19
- # Whisper Tiny - Toki Pona - Synthetic Test 1
20
-
21
- This model is a fine-tuned version of [openai/whisper-tiny](https://huggingface.co/openai/whisper-tiny) on the Common Voice 23.0 - Toki Pona dataset.
22
- It achieves the following results on the evaluation set:
23
- - Loss: 0.1051
24
- - Wer: 6.3891
25
-
26
- ## Model description
27
-
28
- More information needed
29
-
30
- ## Intended uses & limitations
31
-
32
- More information needed
33
-
34
- ## Training and evaluation data
35
-
36
- More information needed
37
-
38
- ## Training procedure
39
-
40
- ### Training hyperparameters
41
-
42
- The following hyperparameters were used during training:
43
- - learning_rate: 1e-05
44
- - train_batch_size: 64
45
- - eval_batch_size: 8
46
- - seed: 42
47
- - gradient_accumulation_steps: 2
48
- - total_train_batch_size: 128
49
- - optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
50
- - lr_scheduler_type: linear
51
- - lr_scheduler_warmup_steps: 100
52
- - training_steps: 1000
53
- - mixed_precision_training: Native AMP
54
-
55
- ### Training results
56
-
57
- | Training Loss | Epoch | Step | Validation Loss | Wer |
58
- |:-------------:|:------:|:----:|:---------------:|:-------:|
59
- | 0.2699 | 0.5155 | 100 | 0.3009 | 16.1318 |
60
- | 0.0985 | 1.0309 | 200 | 0.1711 | 9.6271 |
61
- | 0.0728 | 1.5464 | 300 | 0.1388 | 7.9503 |
62
- | 0.0547 | 2.0619 | 400 | 0.1254 | 7.4299 |
63
- | 0.0474 | 2.5773 | 500 | 0.1178 | 6.6493 |
64
- | 0.039 | 3.0928 | 600 | 0.1128 | 6.5337 |
65
- | 0.0364 | 3.6082 | 700 | 0.1091 | 6.3024 |
66
- | 0.0314 | 4.1237 | 800 | 0.1072 | 6.2735 |
67
- | 0.0304 | 4.6392 | 900 | 0.1059 | 6.3024 |
68
- | 0.0288 | 5.1546 | 1000 | 0.1051 | 6.3891 |
69
-
70
-
71
- ### Framework versions
72
-
73
- - Transformers 4.50.3
74
- - Pytorch 2.9.0+cu126
75
- - Datasets 3.6.0
76
- - Tokenizers 0.21.4
 
1
+ ---
2
+ library_name: transformers
3
+ language:
4
+ - tok
5
+ license: apache-2.0
6
+ base_model: openai/whisper-tiny
7
+ tags:
8
+ - generated_from_trainer
9
+ metrics:
10
+ - wer
11
+ model-index:
12
+ - name: Whisper Tiny - Toki Pona - Synthetic Test 1
13
+ results: []
14
+ ---
15
+
16
+ <!-- This model card has been generated automatically according to the information the Trainer had access to. You
17
+ should probably proofread and complete it, then remove this comment. -->
18
+
19
+ # Whisper Tiny - Toki Pona - Synthetic Test 1
20
+
21
+ This experimental model is a fine-tuned version of [openai/whisper-tiny](https://huggingface.co/openai/whisper-tiny) on a mix of custom synthetic data and Common Voice 23.0 - Toki Pona.
22
+
23
+ The evaluation set contains synthetic data, as opposed to the `whisper-small`-based model, so evaluation values are not provided.
24
+
25
+ ## Model description
26
+
27
+ This is an experimental model trained for speech recognition for Toki Pona.
28
+
29
+ As the original model is multilingual with explicit language specification tokens, we have replaced the Czech (`cs`) language with Toki Pona, as we have determined it to have the closest phonetics.
30
+
31
+ The model's performance for other languages may have been at least partially preserved, but no testing has been done for other languages.
32
+
33
+ ### Training hyperparameters
34
+
35
+ The following hyperparameters were used during training:
36
+ - learning_rate: 1e-05
37
+ - train_batch_size: 64
38
+ - eval_batch_size: 8
39
+ - seed: 42
40
+ - gradient_accumulation_steps: 2
41
+ - total_train_batch_size: 128
42
+ - optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
43
+ - lr_scheduler_type: linear
44
+ - lr_scheduler_warmup_steps: 100
45
+ - training_steps: 1000
46
+ - mixed_precision_training: Native AMP
47
+
48
+ ### Training results
49
+
50
+ | Training Loss | Epoch | Step | Validation Loss | Wer |
51
+ |:-------------:|:------:|:----:|:---------------:|:-------:|
52
+ | 0.2699 | 0.5155 | 100 | 0.3009 | 16.1318 |
53
+ | 0.0985 | 1.0309 | 200 | 0.1711 | 9.6271 |
54
+ | 0.0728 | 1.5464 | 300 | 0.1388 | 7.9503 |
55
+ | 0.0547 | 2.0619 | 400 | 0.1254 | 7.4299 |
56
+ | 0.0474 | 2.5773 | 500 | 0.1178 | 6.6493 |
57
+ | 0.039 | 3.0928 | 600 | 0.1128 | 6.5337 |
58
+ | 0.0364 | 3.6082 | 700 | 0.1091 | 6.3024 |
59
+ | 0.0314 | 4.1237 | 800 | 0.1072 | 6.2735 |
60
+ | 0.0304 | 4.6392 | 900 | 0.1059 | 6.3024 |
61
+ | 0.0288 | 5.1546 | 1000 | 0.1051 | 6.3891 |
62
+
63
+
64
+ ### Framework versions
65
+
66
+ - Transformers 4.50.3
67
+ - Pytorch 2.9.0+cu126
68
+ - Datasets 3.6.0
69
+ - Tokenizers 0.21.4