Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,59 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
language:
|
3 |
+
- de
|
4 |
+
- en
|
5 |
+
license: mit
|
6 |
+
library_name: ctranslate2
|
7 |
+
pipeline_tag: automatic-speech-recognition
|
8 |
+
tags:
|
9 |
+
- whisper-large-v3
|
10 |
+
- ctranslate2
|
11 |
+
- int8_float16
|
12 |
+
- austrian-german
|
13 |
+
- speech-to-text
|
14 |
+
- localai
|
15 |
+
- faster-whisper
|
16 |
+
model_type: whisper
|
17 |
+
base_model: openai/whisper-large-v3
|
18 |
+
quantization: int8_float16
|
19 |
+
inference: true
|
20 |
+
datasets:
|
21 |
+
- common_voice_15_0
|
22 |
+
- mozilla-foundation/common_voice_16_1
|
23 |
+
- own/callcenter_at
|
24 |
+
metrics:
|
25 |
+
- wer
|
26 |
+
---
|
27 |
+
|
28 |
+
# Whisper‐large-v3-DE-AT (CT2 INT8 + FP16)
|
29 |
+
|
30 |
+
Feingetuntes Whisper-large-v3-Modell auf österreichisch-deutschen Sprachdaten.
|
31 |
+
Diese Variante ist **komplett in [CTranslate2]-Format** vorliegend und wurde mit
|
32 |
+
`--quantization int8_float16` exportiert.
|
33 |
+
Dadurch halbiert sich der GPU-VRAM-Bedarf bei praktisch identischer Genauigkeit.
|
34 |
+
|
35 |
+
| Eigenschaft | Wert |
|
36 |
+
|------------------------|--------------------------------------------------|
|
37 |
+
| Basismodell | `openai/whisper-large-v3` |
|
38 |
+
| Quantisierung | **INT8-Gewichte** + FP16-Aktivierungen |
|
39 |
+
| Größe (model.bin) | ≈ 5.6 GB |
|
40 |
+
| Empfohlene Hardware | ≥ 16 GB GPU (VRAM) oder starke CPU mit AVX2/AVX-512 |
|
41 |
+
| Compute Type YAML | `compute_type: int8_float16` |
|
42 |
+
| Licence | MIT (wie Original) |
|
43 |
+
|
44 |
+
---
|
45 |
+
|
46 |
+
## 🛠️ Verwendung
|
47 |
+
|
48 |
+
### LocalAI ≥ v3.0
|
49 |
+
|
50 |
+
```yaml
|
51 |
+
# whisper-large-v3-de-at.yaml
|
52 |
+
name: whisper-large-v3-de-at
|
53 |
+
backend: fast-whisper
|
54 |
+
parameters:
|
55 |
+
model: whisper-large-v3-de-at-ct2
|
56 |
+
language: de
|
57 |
+
translate: false
|
58 |
+
compute_type: int8_float16
|
59 |
+
vad_filter: true
|