agnesluhtaru
commited on
Commit
•
276864c
1
Parent(s):
e9c4c9b
Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,47 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
tags:
|
4 |
+
- generated_from_trainer
|
5 |
+
- whisper-event
|
6 |
+
metrics:
|
7 |
+
- wer
|
8 |
+
---
|
9 |
+
|
10 |
+
# whisper-medium-et with ERR2020 data
|
11 |
+
|
12 |
+
This model is a fine-tuned version of [openai/whisper-medium](https://huggingface.co/openai/whisper-medium) on the following datasets: Common Voice 11, VoxPopuli, FLEURS and [ERR2020](http://bark.phon.ioc.ee/lw/korpused/ERR2020.html).
|
13 |
+
The model is stopped a little early because the Whisper fine-tuning event was ending :)
|
14 |
+
|
15 |
+
## Model description
|
16 |
+
|
17 |
+
More information needed
|
18 |
+
|
19 |
+
## Intended uses & limitations
|
20 |
+
|
21 |
+
More information needed
|
22 |
+
|
23 |
+
## Training and evaluation data
|
24 |
+
|
25 |
+
Estonian data from Common Voice 11, VoxPopuli, FLEURS and ERR2020 corpora as both training and validation sets. Tested on Common Voice 11 test set.
|
26 |
+
|
27 |
+
## Training procedure
|
28 |
+
|
29 |
+
### Training hyperparameters
|
30 |
+
|
31 |
+
The following hyperparameters were used during training:
|
32 |
+
- learning_rate: 1e-05
|
33 |
+
- train_batch_size: 32
|
34 |
+
- eval_batch_size: 16
|
35 |
+
- seed: 42
|
36 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
37 |
+
- lr_scheduler_type: linear
|
38 |
+
- lr_scheduler_warmup_steps: 500
|
39 |
+
- training_steps: 6000
|
40 |
+
- mixed_precision_training: Native AMP
|
41 |
+
|
42 |
+
### Framework versions
|
43 |
+
|
44 |
+
- Transformers 4.26.0.dev0
|
45 |
+
- Pytorch 1.12.1+rocm5.1.1
|
46 |
+
- Datasets 2.7.1.dev0
|
47 |
+
- Tokenizers 0.13.2
|