Commit
•
4dbb38d
1
Parent(s):
d0b9193
update model card README.md
Browse files
README.md
CHANGED
|
@@ -1,10 +1,6 @@
|
|
| 1 |
---
|
| 2 |
-
language:
|
| 3 |
-
- ja
|
| 4 |
license: apache-2.0
|
| 5 |
tags:
|
| 6 |
-
- automatic-speech-recognition
|
| 7 |
-
- mozilla-foundation/common_voice_8_0
|
| 8 |
- generated_from_trainer
|
| 9 |
datasets:
|
| 10 |
- common_voice
|
|
@@ -18,11 +14,11 @@ should probably proofread and complete it, then remove this comment. -->
|
|
| 18 |
|
| 19 |
#
|
| 20 |
|
| 21 |
-
This model is a fine-tuned version of [facebook/wav2vec2-xls-r-1b](https://huggingface.co/facebook/wav2vec2-xls-r-1b) on the
|
| 22 |
It achieves the following results on the evaluation set:
|
| 23 |
-
- Loss: 0.
|
| 24 |
-
- Wer: 1.
|
| 25 |
-
- Cer: 0.
|
| 26 |
|
| 27 |
## Model description
|
| 28 |
|
|
@@ -41,35 +37,25 @@ More information needed
|
|
| 41 |
### Training hyperparameters
|
| 42 |
|
| 43 |
The following hyperparameters were used during training:
|
| 44 |
-
- learning_rate:
|
| 45 |
- train_batch_size: 32
|
| 46 |
- eval_batch_size: 8
|
| 47 |
- seed: 42
|
|
|
|
|
|
|
| 48 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 49 |
- lr_scheduler_type: linear
|
| 50 |
-
- lr_scheduler_warmup_steps:
|
| 51 |
- num_epochs: 50.0
|
| 52 |
- mixed_precision_training: Native AMP
|
| 53 |
|
| 54 |
### Training results
|
| 55 |
|
| 56 |
-
| Training Loss | Epoch | Step
|
| 57 |
-
|
| 58 |
-
| 1.
|
| 59 |
-
|
|
| 60 |
-
|
|
| 61 |
-
| 1.9569 | 12.58 | 4000 | 0.9818 | 1.0090 | 0.2657 |
|
| 62 |
-
| 1.8409 | 15.72 | 5000 | 1.0382 | 1.6480 | 0.3741 |
|
| 63 |
-
| 1.7449 | 18.87 | 6000 | 0.9962 | 1.6268 | 0.3454 |
|
| 64 |
-
| 1.7349 | 22.01 | 7000 | 0.9560 | 0.9850 | 0.2597 |
|
| 65 |
-
| 1.6857 | 25.16 | 8000 | 0.8722 | 0.9669 | 0.2325 |
|
| 66 |
-
| 1.5637 | 28.3 | 9000 | 0.7636 | 1.8071 | 0.3422 |
|
| 67 |
-
| 1.5088 | 31.45 | 10000 | 0.7290 | 1.0398 | 0.2085 |
|
| 68 |
-
| 1.4298 | 34.59 | 11000 | 0.7576 | 1.0166 | 0.2104 |
|
| 69 |
-
| 1.3716 | 37.74 | 12000 | 0.7046 | 1.1275 | 0.2138 |
|
| 70 |
-
| 1.3185 | 40.88 | 13000 | 0.7011 | 1.1696 | 0.2179 |
|
| 71 |
-
| 1.28 | 44.03 | 14000 | 0.6754 | 1.1316 | 0.2024 |
|
| 72 |
-
| 1.2368 | 47.17 | 15000 | 0.6925 | 1.0517 | 0.1923 |
|
| 73 |
|
| 74 |
|
| 75 |
### Framework versions
|
|
|
|
| 1 |
---
|
|
|
|
|
|
|
| 2 |
license: apache-2.0
|
| 3 |
tags:
|
|
|
|
|
|
|
| 4 |
- generated_from_trainer
|
| 5 |
datasets:
|
| 6 |
- common_voice
|
|
|
|
| 14 |
|
| 15 |
#
|
| 16 |
|
| 17 |
+
This model is a fine-tuned version of [facebook/wav2vec2-xls-r-1b](https://huggingface.co/facebook/wav2vec2-xls-r-1b) on the common_voice dataset.
|
| 18 |
It achieves the following results on the evaluation set:
|
| 19 |
+
- Loss: 0.6183
|
| 20 |
+
- Wer: 1.0213
|
| 21 |
+
- Cer: 0.1797
|
| 22 |
|
| 23 |
## Model description
|
| 24 |
|
|
|
|
| 37 |
### Training hyperparameters
|
| 38 |
|
| 39 |
The following hyperparameters were used during training:
|
| 40 |
+
- learning_rate: 7.5e-05
|
| 41 |
- train_batch_size: 32
|
| 42 |
- eval_batch_size: 8
|
| 43 |
- seed: 42
|
| 44 |
+
- gradient_accumulation_steps: 4
|
| 45 |
+
- total_train_batch_size: 128
|
| 46 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 47 |
- lr_scheduler_type: linear
|
| 48 |
+
- lr_scheduler_warmup_steps: 1500
|
| 49 |
- num_epochs: 50.0
|
| 50 |
- mixed_precision_training: Native AMP
|
| 51 |
|
| 52 |
### Training results
|
| 53 |
|
| 54 |
+
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|
| 55 |
+
|:-------------:|:-----:|:----:|:---------------:|:------:|:------:|
|
| 56 |
+
| 1.7019 | 12.65 | 1000 | 1.0510 | 0.9832 | 0.2589 |
|
| 57 |
+
| 1.6385 | 25.31 | 2000 | 0.6670 | 0.9915 | 0.1851 |
|
| 58 |
+
| 1.4344 | 37.97 | 3000 | 0.6183 | 1.0213 | 0.1797 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 59 |
|
| 60 |
|
| 61 |
### Framework versions
|