Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,80 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language: it
|
| 3 |
+
license: mit
|
| 4 |
+
tags:
|
| 5 |
+
- whisper
|
| 6 |
+
- whisper.cpp
|
| 7 |
+
- ggml
|
| 8 |
+
- automatic-speech-recognition
|
| 9 |
+
- italian
|
| 10 |
+
- localai
|
| 11 |
+
- quantized
|
| 12 |
+
datasets:
|
| 13 |
+
- mozilla-foundation/common_voice_25_0
|
| 14 |
+
base_model: openai/whisper-tiny
|
| 15 |
+
pipeline_tag: automatic-speech-recognition
|
| 16 |
+
---
|
| 17 |
+
|
| 18 |
+
# whisper-tiny-it-ggml
|
| 19 |
+
|
| 20 |
+
[whisper.cpp](https://github.com/ggerganov/whisper.cpp) GGML quantizations of [LocalAI-io/whisper-tiny-it](https://huggingface.co/LocalAI-io/whisper-tiny-it) for fast CPU/GPU inference.
|
| 21 |
+
|
| 22 |
+
**Author:** Ettore Di Giacinto
|
| 23 |
+
|
| 24 |
+
Brought to you by the [LocalAI](https://github.com/mudler/LocalAI) team. These models can be used directly with [LocalAI](https://localai.io) and any whisper.cpp-based runtime.
|
| 25 |
+
|
| 26 |
+
## Files
|
| 27 |
+
|
| 28 |
+
| File | Quantization | Description |
|
| 29 |
+
|------|--------------|-------------|
|
| 30 |
+
| `ggml-model-f16.bin` | float16 | Full precision (no quantization) — highest quality |
|
| 31 |
+
| `ggml-model-q8_0.bin` | int8 | 8-bit quantization — minimal quality loss |
|
| 32 |
+
| `ggml-model-q5_1.bin` | int5 | 5-bit quantization with extra precision |
|
| 33 |
+
| `ggml-model-q5_0.bin` | int5 | 5-bit quantization — smallest size |
|
| 34 |
+
|
| 35 |
+
## Training
|
| 36 |
+
|
| 37 |
+
Fine-tuned [openai/whisper-tiny](https://huggingface.co/openai/whisper-tiny) (39M params) on **Common Voice 25.0 Italian**.
|
| 38 |
+
|
| 39 |
+
See [LocalAI-io/whisper-tiny-it](https://huggingface.co/LocalAI-io/whisper-tiny-it) for the full safetensors model and detailed WER results.
|
| 40 |
+
|
| 41 |
+
## Usage
|
| 42 |
+
|
| 43 |
+
### whisper.cpp
|
| 44 |
+
|
| 45 |
+
```bash
|
| 46 |
+
# Download a quant
|
| 47 |
+
huggingface-cli download LocalAI-io/whisper-tiny-it-ggml ggml-model-q5_0.bin --local-dir .
|
| 48 |
+
|
| 49 |
+
# Run
|
| 50 |
+
./whisper-cli -m ggml-model-q5_0.bin -f audio.wav -l it
|
| 51 |
+
```
|
| 52 |
+
|
| 53 |
+
### whisper.cpp Python bindings (pywhispercpp)
|
| 54 |
+
|
| 55 |
+
```python
|
| 56 |
+
from pywhispercpp.model import Model
|
| 57 |
+
|
| 58 |
+
model = Model("ggml-model-q5_0.bin", language="it")
|
| 59 |
+
segments = model.transcribe("audio.wav")
|
| 60 |
+
for seg in segments:
|
| 61 |
+
print(seg.text)
|
| 62 |
+
```
|
| 63 |
+
|
| 64 |
+
### LocalAI
|
| 65 |
+
|
| 66 |
+
```yaml
|
| 67 |
+
# In your LocalAI model config
|
| 68 |
+
name: whisper-tiny-it
|
| 69 |
+
backend: whisper
|
| 70 |
+
parameters:
|
| 71 |
+
model: ggml-model-q5_0.bin
|
| 72 |
+
```
|
| 73 |
+
|
| 74 |
+
## Links
|
| 75 |
+
|
| 76 |
+
- **HF Safetensors:** [LocalAI-io/whisper-tiny-it](https://huggingface.co/LocalAI-io/whisper-tiny-it)
|
| 77 |
+
- **CTranslate2 INT8:** [LocalAI-io/whisper-tiny-it-ct2-int8](https://huggingface.co/LocalAI-io/whisper-tiny-it-ct2-int8)
|
| 78 |
+
- **Code:** [github.com/localai-org/whisper-it](https://github.com/localai-org/whisper-it)
|
| 79 |
+
- **whisper.cpp:** [github.com/ggerganov/whisper.cpp](https://github.com/ggerganov/whisper.cpp)
|
| 80 |
+
- **LocalAI:** [github.com/mudler/LocalAI](https://github.com/mudler/LocalAI)
|