waveletdeboshir
commited on
Commit
•
b400cb5
1
Parent(s):
a33e071
Update README.md
Browse files
README.md
CHANGED
@@ -32,19 +32,6 @@ Model size is 15% less then original whisper-small:
|
|
32 |
| model file size | 967 Mb | 837 Mb |
|
33 |
| vocab_size | 51865 | 4705 |
|
34 |
|
35 |
-
## Other pruned whisper models
|
36 |
-
* [waveletdeboshir/whisper-tiny-ru-pruned](https://huggingface.co/waveletdeboshir/whisper-tiny-ru-pruned)
|
37 |
-
* [waveletdeboshir/whisper-base-ru-pruned](https://huggingface.co/waveletdeboshir/whisper-base-ru-pruned)
|
38 |
-
|
39 |
-
## Metrics
|
40 |
-
| | openai/whisper-small | waveletdeboshir/whisper-small-ru-pruned |
|
41 |
-
| :------ | :------ | :------ |
|
42 |
-
| WER* golos-test-crowd | 0.3358 | 0.3471 |
|
43 |
-
| CER* golos-test-crowd | 0.1561 | 0.1444 |
|
44 |
-
*Metrics were measured after text normalization
|
45 |
-
|
46 |
-
You can fine-tune this model on your data to achive better performance.
|
47 |
-
|
48 |
## Usage
|
49 |
Model can be used as an original whisper:
|
50 |
|
@@ -70,5 +57,18 @@ Model can be used as an original whisper:
|
|
70 |
```
|
71 |
The context tokens can be removed from the start of the transcription by setting `skip_special_tokens=True`.
|
72 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
73 |
## Colab for pruning
|
74 |
TODO
|
|
|
32 |
| model file size | 967 Mb | 837 Mb |
|
33 |
| vocab_size | 51865 | 4705 |
|
34 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
35 |
## Usage
|
36 |
Model can be used as an original whisper:
|
37 |
|
|
|
57 |
```
|
58 |
The context tokens can be removed from the start of the transcription by setting `skip_special_tokens=True`.
|
59 |
|
60 |
+
## Other pruned whisper models
|
61 |
+
* [waveletdeboshir/whisper-tiny-ru-pruned](https://huggingface.co/waveletdeboshir/whisper-tiny-ru-pruned)
|
62 |
+
* [waveletdeboshir/whisper-base-ru-pruned](https://huggingface.co/waveletdeboshir/whisper-base-ru-pruned)
|
63 |
+
|
64 |
+
## Metrics
|
65 |
+
| | openai/whisper-small | waveletdeboshir/whisper-small-ru-pruned |
|
66 |
+
| :------ | :------ | :------ |
|
67 |
+
| WER* golos-test-crowd | 0.3358 | 0.3471 |
|
68 |
+
| CER* golos-test-crowd | 0.1561 | 0.1444 |
|
69 |
+
*Metrics were measured after text normalization
|
70 |
+
|
71 |
+
You can fine-tune this model on your data to achive better performance.
|
72 |
+
|
73 |
## Colab for pruning
|
74 |
TODO
|