End of training
Browse files
README.md
CHANGED
|
@@ -5,9 +5,22 @@ tags:
|
|
| 5 |
- generated_from_trainer
|
| 6 |
datasets:
|
| 7 |
- mozilla-foundation/common_voice_6_1
|
|
|
|
|
|
|
| 8 |
model-index:
|
| 9 |
- name: Whisper Small Frisian 10h
|
| 10 |
-
results:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
---
|
| 12 |
|
| 13 |
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
|
@@ -17,13 +30,8 @@ should probably proofread and complete it, then remove this comment. -->
|
|
| 17 |
|
| 18 |
This model is a fine-tuned version of [openai/whisper-small](https://huggingface.co/openai/whisper-small) on the Common Voice 6.1 dataset.
|
| 19 |
It achieves the following results on the evaluation set:
|
| 20 |
-
-
|
| 21 |
-
-
|
| 22 |
-
- eval_runtime: 599.1927
|
| 23 |
-
- eval_samples_per_second: 5.04
|
| 24 |
-
- eval_steps_per_second: 0.631
|
| 25 |
-
- epoch: 1.0695
|
| 26 |
-
- step: 1000
|
| 27 |
|
| 28 |
## Model description
|
| 29 |
|
|
@@ -49,9 +57,30 @@ The following hyperparameters were used during training:
|
|
| 49 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 50 |
- lr_scheduler_type: linear
|
| 51 |
- lr_scheduler_warmup_steps: 50
|
| 52 |
-
- training_steps:
|
| 53 |
- mixed_precision_training: Native AMP
|
| 54 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 55 |
### Framework versions
|
| 56 |
|
| 57 |
- Transformers 4.40.1
|
|
|
|
| 5 |
- generated_from_trainer
|
| 6 |
datasets:
|
| 7 |
- mozilla-foundation/common_voice_6_1
|
| 8 |
+
metrics:
|
| 9 |
+
- wer
|
| 10 |
model-index:
|
| 11 |
- name: Whisper Small Frisian 10h
|
| 12 |
+
results:
|
| 13 |
+
- task:
|
| 14 |
+
name: Automatic Speech Recognition
|
| 15 |
+
type: automatic-speech-recognition
|
| 16 |
+
dataset:
|
| 17 |
+
name: Common Voice 6.1
|
| 18 |
+
type: mozilla-foundation/common_voice_6_1
|
| 19 |
+
args: 'config: frisian, split: test'
|
| 20 |
+
metrics:
|
| 21 |
+
- name: Wer
|
| 22 |
+
type: wer
|
| 23 |
+
value: 22.427374799500978
|
| 24 |
---
|
| 25 |
|
| 26 |
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
|
|
|
| 30 |
|
| 31 |
This model is a fine-tuned version of [openai/whisper-small](https://huggingface.co/openai/whisper-small) on the Common Voice 6.1 dataset.
|
| 32 |
It achieves the following results on the evaluation set:
|
| 33 |
+
- Loss: 0.3402
|
| 34 |
+
- Wer: 22.4274
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 35 |
|
| 36 |
## Model description
|
| 37 |
|
|
|
|
| 57 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 58 |
- lr_scheduler_type: linear
|
| 59 |
- lr_scheduler_warmup_steps: 50
|
| 60 |
+
- training_steps: 1500
|
| 61 |
- mixed_precision_training: Native AMP
|
| 62 |
|
| 63 |
+
### Training results
|
| 64 |
+
|
| 65 |
+
| Training Loss | Epoch | Step | Validation Loss | Wer |
|
| 66 |
+
|:-------------:|:------:|:----:|:---------------:|:-------:|
|
| 67 |
+
| 1.0548 | 0.1070 | 100 | 1.0200 | 52.0620 |
|
| 68 |
+
| 0.6944 | 0.2139 | 200 | 0.7126 | 39.6222 |
|
| 69 |
+
| 0.6024 | 0.3209 | 300 | 0.6052 | 36.0791 |
|
| 70 |
+
| 0.4697 | 0.4278 | 400 | 0.5303 | 32.5040 |
|
| 71 |
+
| 0.4222 | 0.5348 | 500 | 0.4780 | 30.0766 |
|
| 72 |
+
| 0.4075 | 0.6417 | 600 | 0.4458 | 28.4691 |
|
| 73 |
+
| 0.374 | 0.7487 | 700 | 0.4151 | 26.9292 |
|
| 74 |
+
| 0.3381 | 0.8556 | 800 | 0.3949 | 25.4678 |
|
| 75 |
+
| 0.3235 | 0.9626 | 900 | 0.3764 | 24.8904 |
|
| 76 |
+
| 0.1861 | 1.0695 | 1000 | 0.3643 | 23.5716 |
|
| 77 |
+
| 0.1554 | 1.1765 | 1100 | 0.3608 | 23.4183 |
|
| 78 |
+
| 0.1639 | 1.2834 | 1200 | 0.3511 | 23.0298 |
|
| 79 |
+
| 0.1453 | 1.3904 | 1300 | 0.3449 | 22.6591 |
|
| 80 |
+
| 0.1531 | 1.4973 | 1400 | 0.3419 | 22.4452 |
|
| 81 |
+
| 0.1299 | 1.6043 | 1500 | 0.3402 | 22.4274 |
|
| 82 |
+
|
| 83 |
+
|
| 84 |
### Framework versions
|
| 85 |
|
| 86 |
- Transformers 4.40.1
|