README.md

metadata

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 as whisper.cpp runs it, mirrored unchanged from 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.