File size: 2,718 Bytes
e1f45a7 0ecb6f5 e1f45a7 | 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 | ---
license: mit
base_model: openai/whisper-tiny
library_name: whisper.cpp
pipeline_tag: automatic-speech-recognition
tags:
- whisper
- whisper.cpp
- ggml
- coreml
- ios
---
# Whisper Tiny — whisper.cpp (ggml + Core ML encoder)
OpenAI's [Whisper tiny](https://huggingface.co/openai/whisper-tiny) as whisper.cpp runs it, mirrored unchanged from [ggerganov/whisper.cpp](https://huggingface.co/ggerganov/whisper.cpp) at revision `5359861c739e955e79d9a303bcbc70fb988958b1`, for on-device transcription in the Character Agent iOS app. Not affiliated with or endorsed by OpenAI or the whisper.cpp authors.
## Files
- `ggml-tiny-q5_1.bin`: Whisper Tiny in ggml format, weights quantized (Q5_1); `ggml-tiny-encoder.mlmodelc.zip`: its encoder compiled for Core ML (fp16), for the Apple Neural Engine.
- `ggml-tiny.en-q5_1.bin`: Whisper Tiny (English only) in ggml format, weights quantized (Q5_1); `ggml-tiny.en-encoder.mlmodelc.zip`: its encoder compiled for Core ML (fp16), for the Apple Neural Engine.
Unzip an encoder next to its `.bin`: whisper.cpp built with Core ML finds `ggml-<model>-encoder.mlmodelc` by dropping the quantization suffix, and falls back to its own encoder if it's missing.
## Files and SHA-256
| File | Bytes | SHA-256 |
|---|---|---|
| `ggml-tiny-q5_1.bin` | 32,152,673 | `818710568da3ca15689e31a743197b520007872ff9576237bda97bd1b469c3d7` |
| `ggml-tiny-encoder.mlmodelc.zip` | 15,037,446 | `c88cbd2648e1f5415092bcf5256add463a0f19943e6938f46e8d4ffdebd47739` |
| `ggml-tiny.en-q5_1.bin` | 32,166,155 | `c77c5766f1cef09b6b7d47f21b546cbddd4157886b3b5d6d4f709e91e66c7c2b` |
| `ggml-tiny.en-encoder.mlmodelc.zip` | 15,034,655 | `82b32eef73c94bb0c432a776a047b757d9525c26d84038a15d8798d7c8d1ee58` |
The files above are byte-identical to the files at the revision above.
## Decoder-only files
- `ggml-tiny-q5_1-decoder.bin`: `ggml-tiny-q5_1.bin` with the encoder's tensors (`encoder.*`) removed, for whisper.cpp builds that run the encoder only through its Core ML encoder (Character Agent patches whisper.cpp to load these).
- `ggml-tiny.en-q5_1-decoder.bin`: `ggml-tiny.en-q5_1.bin` with the encoder's tensors (`encoder.*`) removed, for whisper.cpp builds that run the encoder only through its Core ML encoder (Character Agent patches whisper.cpp to load these).
| File | Bytes | SHA-256 |
|---|---|---|
| `ggml-tiny-q5_1-decoder.bin` | 23,388,163 | `8e892e9a3d7e8857543a07aa6197ab98131f069e9f46ecef15381cd4fca77200` |
| `ggml-tiny.en-q5_1-decoder.bin` | 23,401,645 | `38190ac024471999a5830d8fb05d22951ed218323098b5c34b60d67f39526e60` |
## License
MIT, the same as Whisper and whisper.cpp (see `LICENSE`). Model © OpenAI; ggml conversion and Core ML encoders © the ggml authors.
|