sfabriece commited on
Commit
3b7555e
·
verified ·
1 Parent(s): 28b799c

Upload folder using huggingface_hub

Browse files
Files changed (3) hide show
  1. README.md +56 -0
  2. config.json +13 -0
  3. weights.safetensors +3 -0
README.md ADDED
@@ -0,0 +1,56 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ base_model: jlvdoorn/whisper-large-v3-atco2-asr
4
+ library_name: mlx
5
+ tags:
6
+ - whisper
7
+ - mlx
8
+ - apple-silicon
9
+ - air-traffic-control
10
+ - atc
11
+ - aviation
12
+ - automatic-speech-recognition
13
+ pipeline_tag: automatic-speech-recognition
14
+ language:
15
+ - en
16
+ ---
17
+
18
+ # whisper-large-v3-atco2-asr-mlx
19
+
20
+ MLX (float16) build of [`jlvdoorn/whisper-large-v3-atco2-asr`](https://huggingface.co/jlvdoorn/whisper-large-v3-atco2-asr) for Apple Silicon inference via [`mlx-whisper`](https://github.com/ml-explore/mlx-examples/tree/main/whisper).
21
+
22
+ The base model is an aviation-fine-tuned Whisper-large-v3 that reports ~17% WER on ATCO2 vs ~37–55% WER for stock large-v3 on the same domain. No public MLX conversion existed at conversion time (2026-04), so this repo exists to unblock MPS evaluation.
23
+
24
+ ## Usage
25
+
26
+ ```python
27
+ import mlx_whisper
28
+
29
+ result = mlx_whisper.transcribe(
30
+ "audio.wav",
31
+ path_or_hf_repo="sfabriece/whisper-large-v3-atco2-asr-mlx",
32
+ language="en",
33
+ )
34
+ print(result["text"])
35
+ ```
36
+
37
+ ## Conversion
38
+
39
+ Converted with the `mlx-examples/whisper/convert.py` script:
40
+
41
+ ```bash
42
+ python convert.py \
43
+ --torch-name-or-path jlvdoorn/whisper-large-v3-atco2-asr \
44
+ --mlx-path ./whisper-large-v3-atco2-asr-mlx \
45
+ --dtype float16
46
+ # then rename model.safetensors → weights.safetensors (mlx-whisper loader expects this name)
47
+ ```
48
+
49
+ ## License
50
+
51
+ Apache-2.0 (inherited from the base model).
52
+
53
+ ## Attribution
54
+
55
+ - Base model: [`jlvdoorn/whisper-large-v3-atco2-asr`](https://huggingface.co/jlvdoorn/whisper-large-v3-atco2-asr) — aviation fine-tune on top of OpenAI Whisper large-v3 (MIT).
56
+ - Conversion tooling: [`ml-explore/mlx-examples/whisper`](https://github.com/ml-explore/mlx-examples/tree/main/whisper).
config.json ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "n_mels": 128,
3
+ "n_audio_ctx": 1500,
4
+ "n_audio_state": 1280,
5
+ "n_audio_head": 20,
6
+ "n_audio_layer": 32,
7
+ "n_vocab": 51866,
8
+ "n_text_ctx": 448,
9
+ "n_text_state": 1280,
10
+ "n_text_head": 20,
11
+ "n_text_layer": 32,
12
+ "model_type": "whisper"
13
+ }
weights.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9b56f9783729dad8a8b4d77c9ed69910407cf3028820c9e4787eaa6b93cceed4
3
+ size 3083280590