Update README.md
Browse files
README.md
CHANGED
@@ -63,7 +63,7 @@ It achieves the following results:
|
|
63 |
- WER on ShEMO dev set: 32.85
|
64 |
- WER on Common Voice 13 test set: 19.21
|
65 |
|
66 |
-
## Evaluation results
|
67 |
| Checkpoint Name | WER on ShEMO dev set | WER on Common Voice 13 test set | Max :) |
|
68 |
| :---------------------------------------------------------------------------------------------------------------: | :------: | :-------: | :---: |
|
69 |
| [m3hrdadfi/wav2vec2-large-xlsr-persian-v3](https://huggingface.co/m3hrdadfi/wav2vec2-large-xlsr-persian-v3) | 46.55 | **17.43** | 46.55 |
|
@@ -73,7 +73,7 @@ It achieves the following results:
|
|
73 |
|
74 |
As you can see, my model performs better in maximum case :D
|
75 |
|
76 |
-
## Training procedure
|
77 |
|
78 |
#### Training hyperparameters
|
79 |
|
@@ -92,7 +92,7 @@ The following hyperparameters were used during training:
|
|
92 |
|
93 |
You may need *gradient_accumulation* because you need more batch size.
|
94 |
|
95 |
-
#### Training
|
96 |
|
97 |
| Training Loss | Epoch | Step | Validation Loss | Wer |
|
98 |
|:-------------:|:-----:|:----:|:---------------:|:------:|
|
@@ -117,7 +117,7 @@ You may need *gradient_accumulation* because you need more batch size.
|
|
117 |
| 0.8238 | 11.88 | 1900 | 0.6735 | 0.3297 |
|
118 |
| 0.7618 | 12.5 | 2000 | 0.6728 | 0.3286 |
|
119 |
|
120 |
-
#### Hyperparameter tuning
|
121 |
Several models with differet hyperparameters were trained. The following figures show the training process for three of them.
|
122 |
![wer](wandb-wer.png)
|
123 |
![loss](wandb-loss.png)
|
@@ -182,5 +182,7 @@ Check out [this blog](https://huggingface.co/blog/fine-tune-xlsr-wav2vec2) for m
|
|
182 |
## Contact us 🤝
|
183 |
If you have any technical question regarding the model, pretraining, code or publication, please create an issue in the repository. This is the *best* way to reach us.
|
184 |
|
185 |
-
## Citation
|
186 |
-
*TO DO!*
|
|
|
|
|
|
63 |
- WER on ShEMO dev set: 32.85
|
64 |
- WER on Common Voice 13 test set: 19.21
|
65 |
|
66 |
+
## Evaluation results 🧪
|
67 |
| Checkpoint Name | WER on ShEMO dev set | WER on Common Voice 13 test set | Max :) |
|
68 |
| :---------------------------------------------------------------------------------------------------------------: | :------: | :-------: | :---: |
|
69 |
| [m3hrdadfi/wav2vec2-large-xlsr-persian-v3](https://huggingface.co/m3hrdadfi/wav2vec2-large-xlsr-persian-v3) | 46.55 | **17.43** | 46.55 |
|
|
|
73 |
|
74 |
As you can see, my model performs better in maximum case :D
|
75 |
|
76 |
+
## Training procedure 🏋️
|
77 |
|
78 |
#### Training hyperparameters
|
79 |
|
|
|
92 |
|
93 |
You may need *gradient_accumulation* because you need more batch size.
|
94 |
|
95 |
+
#### Training log
|
96 |
|
97 |
| Training Loss | Epoch | Step | Validation Loss | Wer |
|
98 |
|:-------------:|:-----:|:----:|:---------------:|:------:|
|
|
|
117 |
| 0.8238 | 11.88 | 1900 | 0.6735 | 0.3297 |
|
118 |
| 0.7618 | 12.5 | 2000 | 0.6728 | 0.3286 |
|
119 |
|
120 |
+
#### Hyperparameter tuning 🔧
|
121 |
Several models with differet hyperparameters were trained. The following figures show the training process for three of them.
|
122 |
![wer](wandb-wer.png)
|
123 |
![loss](wandb-loss.png)
|
|
|
182 |
## Contact us 🤝
|
183 |
If you have any technical question regarding the model, pretraining, code or publication, please create an issue in the repository. This is the *best* way to reach us.
|
184 |
|
185 |
+
## Citation 🖇
|
186 |
+
*TO DO!*
|
187 |
+
|
188 |
+
**Fine-tuned with ❤️ without ☕︎**
|