End of training
Browse files
README.md
CHANGED
|
@@ -20,8 +20,8 @@ should probably proofread and complete it, then remove this comment. -->
|
|
| 20 |
|
| 21 |
This model is a fine-tuned version of [openai/whisper-small](https://huggingface.co/openai/whisper-small) on the Common Voice 11.0 dataset.
|
| 22 |
It achieves the following results on the evaluation set:
|
| 23 |
-
- Loss:
|
| 24 |
-
- Wer:
|
| 25 |
|
| 26 |
## Model description
|
| 27 |
|
|
@@ -47,33 +47,14 @@ The following hyperparameters were used during training:
|
|
| 47 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 48 |
- lr_scheduler_type: linear
|
| 49 |
- lr_scheduler_warmup_steps: 500
|
| 50 |
-
- training_steps:
|
| 51 |
- mixed_precision_training: Native AMP
|
| 52 |
|
| 53 |
### Training results
|
| 54 |
|
| 55 |
| Training Loss | Epoch | Step | Validation Loss | Wer |
|
| 56 |
|:-------------:|:------:|:----:|:---------------:|:--------:|
|
| 57 |
-
| 3.
|
| 58 |
-
| 2.413 | 0.2232 | 50 | 2.2669 | 202.0585 |
|
| 59 |
-
| 1.8868 | 0.3348 | 75 | 1.5418 | 121.6145 |
|
| 60 |
-
| 0.9929 | 0.4464 | 100 | 0.9056 | 56.4279 |
|
| 61 |
-
| 0.6453 | 0.5580 | 125 | 0.5584 | 46.2059 |
|
| 62 |
-
| 0.3948 | 0.6696 | 150 | 0.3752 | 32.2704 |
|
| 63 |
-
| 0.2766 | 0.7812 | 175 | 0.2828 | 22.4117 |
|
| 64 |
-
| 0.2132 | 0.8929 | 200 | 0.2279 | 15.1665 |
|
| 65 |
-
| 0.2156 | 1.0045 | 225 | 0.1959 | 12.5328 |
|
| 66 |
-
| 0.1292 | 1.1161 | 250 | 0.1718 | 10.8174 |
|
| 67 |
-
| 0.1257 | 1.2277 | 275 | 0.1581 | 7.9717 |
|
| 68 |
-
| 0.169 | 1.3393 | 300 | 0.1485 | 8.0626 |
|
| 69 |
-
| 0.1276 | 1.4509 | 325 | 0.1393 | 7.8910 |
|
| 70 |
-
| 0.1012 | 1.5625 | 350 | 0.1305 | 6.3169 |
|
| 71 |
-
| 0.108 | 1.6741 | 375 | 0.1264 | 8.0827 |
|
| 72 |
-
| 0.0979 | 1.7857 | 400 | 0.1207 | 6.5388 |
|
| 73 |
-
| 0.079 | 1.8973 | 425 | 0.1147 | 5.9939 |
|
| 74 |
-
| 0.0713 | 2.0089 | 450 | 0.1080 | 6.8416 |
|
| 75 |
-
| 0.039 | 2.1205 | 475 | 0.1110 | 5.2775 |
|
| 76 |
-
| 0.0482 | 2.2321 | 500 | 0.1111 | 4.7225 |
|
| 77 |
|
| 78 |
|
| 79 |
### Framework versions
|
|
|
|
| 20 |
|
| 21 |
This model is a fine-tuned version of [openai/whisper-small](https://huggingface.co/openai/whisper-small) on the Common Voice 11.0 dataset.
|
| 22 |
It achieves the following results on the evaluation set:
|
| 23 |
+
- Loss: 3.1977
|
| 24 |
+
- Wer: 269.0515
|
| 25 |
|
| 26 |
## Model description
|
| 27 |
|
|
|
|
| 47 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 48 |
- lr_scheduler_type: linear
|
| 49 |
- lr_scheduler_warmup_steps: 500
|
| 50 |
+
- training_steps: 5
|
| 51 |
- mixed_precision_training: Native AMP
|
| 52 |
|
| 53 |
### Training results
|
| 54 |
|
| 55 |
| Training Loss | Epoch | Step | Validation Loss | Wer |
|
| 56 |
|:-------------:|:------:|:----:|:---------------:|:--------:|
|
| 57 |
+
| 3.2134 | 0.0223 | 5 | 3.1977 | 269.0515 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 58 |
|
| 59 |
|
| 60 |
### Framework versions
|