File size: 1,467 Bytes
9973009
 
 
 
98ae8a9
9973009
 
 
666fdce
9973009
 
 
40fc444
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
---
language: ml
library_name: mlx
license: apache-2.0
pipeline_tag: automatic-speech-recognition
tags:
- whisper
- mlx
- Automatic Speech Recognition
base_model: vrclc/Whisper-medium-Malayalam
---

## Whisper-medium-Malayalam (MLX)

Apple MLX-converted weights for `vrclc/Whisper-medium-Malayalam` optimized for Apple Silicon.

- Base model: `vrclc/Whisper-medium-Malayalam`
- Format: MLX (`weights.safetensors`, `config.json`)
- Intended runtime: `mlx-whisper` on Apple Silicon (M-series)

### Usage (Python)
```python
import mlx_whisper

result = mlx_whisper.transcribe(
    "/path/to/audio.wav",
    path_or_hf_repo="<this-repo>",
    # Optional decoding controls
    language="ml",               # Malayalam
    task="transcribe",           # or "translate"
    temperature=0.0,
    no_speech_threshold=0.3,
    logprob_threshold=-1.0,
    compression_ratio_threshold=2.4,
)
print(result["text"]) 
```

### Local HTTP server (FastAPI)
With the server at `whisper/server_mlx.py` from `avatar-npm`:
```bash
export WHISPER_MODEL=<this-repo-or-local-mlx-path>
export WHISPER_LANGUAGE=ml
python server_mlx.py
# POST /transcribe with form field `file`
```

### Notes
- This repo contains only the MLX weights and config. Tokenization and audio
  preprocessing are handled by `mlx-whisper`.
- If you need the original (non-MLX) model, see `vrclc/Whisper-medium-Malayalam`.

### License
The original model’s license applies. See the upstream repository for details.