File size: 1,506 Bytes
6ebccf5
6768b90
 
6ebccf5
6768b90
 
6ebccf5
 
6768b90
 
6ebccf5
 
 
 
 
6768b90
6ebccf5
 
 
 
 
 
6768b90
 
 
6ebccf5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
---
language:
- ar
license: apache-2.0
datasets:
- mozilla-foundation/common_voice_13_0
metrics:
- wer
pipeline_tag: automatic-speech-recognition
base_model: openai/whisper-small
model-index:
- name: whisper_small_hi_flax
  results:
  - task:
      type: automatic-speech-recognition
      name: Automatic Speech Recognition
    dataset:
      name: Common Voice 13.0
      type: mozilla-foundation/common_voice_13_0
      config: hi
      split: test
    metrics:
    - type: wer
      value: 33.96828
      name: Wer
---

# Whisper Small Hi - Sanchit Gandhi

This model is a fine-tuned version of [openai/whisper-small](https://huggingface.co/openai/whisper-small) on the Common Voice 13.0 dataset in Flax.
It is trained using the Transformers **Flax** examples script, and achieves the following results on the evaluation set:
- Loss: 0.02091
- Wer: 33.96828

The training run can be reproduced in approximately 25 minutes by executing the script [`run.sh`](https://huggingface.co/sanchit-gandhi/whisper-small-hi-flax/blob/main/run.sh).

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 1e-04
- train_batch_size: 64
- eval_batch_size: 64
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_train_epochs: 10

### Training results

See [Tensorboard logs](https://huggingface.co/sanchit-gandhi/whisper-small-hi-flax/tensorboard) for details.