Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,84 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
- pl
|
| 6 |
+
- de
|
| 7 |
+
- fr
|
| 8 |
+
- es
|
| 9 |
+
- it
|
| 10 |
+
- pt
|
| 11 |
+
- nl
|
| 12 |
+
- ru
|
| 13 |
+
- zh
|
| 14 |
+
- ja
|
| 15 |
+
- ko
|
| 16 |
+
- ar
|
| 17 |
+
- hi
|
| 18 |
+
- multilingual
|
| 19 |
+
tags:
|
| 20 |
+
- whisper
|
| 21 |
+
- speech-to-text
|
| 22 |
+
- mlx
|
| 23 |
+
- quantized
|
| 24 |
+
- q4
|
| 25 |
+
- 4bit
|
| 26 |
+
- apple-silicon
|
| 27 |
+
library_name: transformers
|
| 28 |
+
pipeline_tag: automatic-speech-recognition
|
| 29 |
+
base_model: openai/whisper-large-v3
|
| 30 |
+
---
|
| 31 |
+
|
| 32 |
+
# Whisper Large V3 - MLX Q4 Quantized
|
| 33 |
+
|
| 34 |
+
4-bit quantized version of OpenAI's Whisper Large V3, optimized for Apple Silicon with MLX. Smallest large-v3 variant.
|
| 35 |
+
|
| 36 |
+
## Model Details
|
| 37 |
+
|
| 38 |
+
| Property | Value |
|
| 39 |
+
|----------|-------|
|
| 40 |
+
| Original Model | [openai/whisper-large-v3](https://huggingface.co/openai/whisper-large-v3) |
|
| 41 |
+
| Parameters | ~1.55B |
|
| 42 |
+
| Quantization | INT4 (Q4) |
|
| 43 |
+
| Size | ~900MB |
|
| 44 |
+
| Decoder Layers | 32 |
|
| 45 |
+
|
| 46 |
+
Half the size of Q8 with minimal accuracy loss. Best choice for memory-constrained devices.
|
| 47 |
+
|
| 48 |
+
## Other Whisper Models
|
| 49 |
+
|
| 50 |
+
| Model | Size | Quality | Link |
|
| 51 |
+
|-------|------|---------|------|
|
| 52 |
+
| small | ~300MB | Good | [LibraxisAI/whisper-small-mlx-q8](https://huggingface.co/LibraxisAI/whisper-small-mlx-q8) |
|
| 53 |
+
| medium | ~800MB | Better | [LibraxisAI/whisper-medium-mlx-q8](https://huggingface.co/LibraxisAI/whisper-medium-mlx-q8) |
|
| 54 |
+
| large-v3 q8 | ~1.6GB | Best | [LibraxisAI/whisper-large-v3-mlx-q8](https://huggingface.co/LibraxisAI/whisper-large-v3-mlx-q8) |
|
| 55 |
+
| **large-v3 q4** (this) | ~900MB | Best (compressed) | - |
|
| 56 |
+
| large-v3-turbo | ~900MB | Great (fast) | [LibraxisAI/whisper-large-v3-turbo-mlx-q8](https://huggingface.co/LibraxisAI/whisper-large-v3-turbo-mlx-q8) |
|
| 57 |
+
|
| 58 |
+
## Usage
|
| 59 |
+
|
| 60 |
+
```python
|
| 61 |
+
import mlx_whisper
|
| 62 |
+
|
| 63 |
+
result = mlx_whisper.transcribe(
|
| 64 |
+
"audio.wav",
|
| 65 |
+
path_or_hf_repo="LibraxisAI/whisper-large-v3-q4"
|
| 66 |
+
)
|
| 67 |
+
print(result["text"])
|
| 68 |
+
```
|
| 69 |
+
|
| 70 |
+
## Supported Languages
|
| 71 |
+
|
| 72 |
+
Full multilingual support: English, Polish, German, French, Spanish, Italian, Portuguese, Dutch, Russian, Chinese, Japanese, Korean, Arabic, Hindi, and 90+ additional languages.
|
| 73 |
+
|
| 74 |
+
## Hardware Requirements
|
| 75 |
+
|
| 76 |
+
- Apple Silicon Mac (M1/M2/M3/M4)
|
| 77 |
+
- Minimum 8GB RAM
|
| 78 |
+
|
| 79 |
+
## License
|
| 80 |
+
|
| 81 |
+
MIT - inherited from OpenAI Whisper.
|
| 82 |
+
|
| 83 |
+
---
|
| 84 |
+
*Converted by [LibraxisAI](https://huggingface.co/LibraxisAI) using mlx-whisper*
|