waveletdeboshir
commited on
Commit
•
a8a91ce
1
Parent(s):
b083b1e
Add info about model
Browse files
README.md
CHANGED
@@ -1,3 +1,42 @@
|
|
1 |
-
---
|
2 |
-
license: apache-2.0
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
language:
|
4 |
+
- ru
|
5 |
+
library_name: transformers
|
6 |
+
pipeline_tag: automatic-speech-recognition
|
7 |
+
tags:
|
8 |
+
- asr
|
9 |
+
- Pytorch
|
10 |
+
- pruned
|
11 |
+
- audio
|
12 |
+
- automatic-speech-recognition
|
13 |
+
---
|
14 |
+
|
15 |
+
# Whisper-tiny-ru-pruned
|
16 |
+
|
17 |
+
## Model info
|
18 |
+
This is a pruned version of [openai/whisper-tiny](https://huggingface.co/openai/whisper-tiny) model with only russian tokens left.
|
19 |
+
Pruning was made without any fine-tuning. Method from [this post](https://medium.com/m/global-identity-2?redirectUrl=https%3A%2F%2Ftowardsdatascience.com%2Fhow-to-adapt-a-multilingual-t5-model-for-a-single-language-b9f94f3d9c90) was used.
|
20 |
+
|
21 |
+
## Size
|
22 |
+
Only 10% tokens was left including special whisper tokens, added whisper tokens, 100 most popular tokens from tokenizer and 3000 most popular Russian tokens computed by tokenization of russian text corpus.
|
23 |
+
|
24 |
+
Model size is 50% less then original whisper-small:
|
25 |
+
| | openai/whisper-tiny | waveletdeboshir/whisper-tiny-ru-pruned |
|
26 |
+
| :------ | :------ | :------ |
|
27 |
+
| n of parameters | 38 M | 19.6 M |
|
28 |
+
| n of parameters (with proj_out layer) | 57.6 M | 21.5 M |
|
29 |
+
| model file size | 151 Mb | 86 Mb |
|
30 |
+
| vocab_size | 51865 | 4705 |
|
31 |
+
|
32 |
+
## Other pruned whisper models
|
33 |
+
[waveletdeboshir/whisper-base-ru-pruned](https://huggingface.co/waveletdeboshir/whisper-base-ru-pruned)
|
34 |
+
[waveletdeboshir/whisper-small-ru-pruned](https://huggingface.co/waveletdeboshir/whisper-small-ru-pruned)
|
35 |
+
|
36 |
+
## Metrics
|
37 |
+
TODO
|
38 |
+
|
39 |
+
You can fine-tune this model on your data to achive better performance.
|
40 |
+
|
41 |
+
## Colab for pruning
|
42 |
+
TODO
|