Whisper Small: Core ML encoder, GGUF decoders and full models

This repository contains the optional Apple Silicon package for Whisper Small in Glimpse: a compiled Core ML encoder and decoder-only GGUFs. A decoder-only GGUF carries the decoder, tokenizer and metadata without the encoder weights, so the encoder is not downloaded twice. The encoder also works beside a full Whisper GGUF or whisper.cpp ggml model. Glimpse uses the encoder for every Whisper Small quantization.

File	Download size	Purpose
`whisper-small-encoder.mlmodelc.zip`	163.1 MB	Compiled Core ML encoder
`whisper-small-Q8_0-decoder.gguf`	166.6 MB	Decoder, tokenizer and metadata of `whisper-small-Q8_0.gguf`

A decoder-only GGUF needs the encoder. On Windows, Intel Macs, or anywhere without Core ML, use the full model file instead. The Core ML package requires Apple Silicon.

Full models

Complete model files for Windows, Intel Macs and any CPU/GPU path. They need no encoder.

File	Download size	Format
`whisper-small-Q8_0.gguf`	269.8 MB	Full GGUF, Q8_0
`whisper-small-Q5_K_M.gguf`	193.7 MB	Full GGUF, Q5_K_M
`whisper-small-F16.gguf`	492.9 MB	Full GGUF, F16
`ggml-small-q5_1.bin`	190.1 MB	whisper.cpp ggml, Q5_1
`ggml-small-q8_0.bin`	264.5 MB	whisper.cpp ggml, Q8_0
`ggml-small.bin`	487.6 MB	whisper.cpp ggml, F16

Provenance

The model originates from OpenAI Whisper Small, licensed Apache 2.0. The encoder was converted from whisper.cpp's F16 ggml-small.bin with scripts/convert-whisper-gguf-to-coreml.py from our transcribe.cpp fork, then zipped with ditto -c -k --keepParent. Each decoder-only GGUF comes from the full GGUF listed below via scripts/extract-coreml-decoder.py, which copies the decoder and frontend tensors and metadata byte for byte, omits the encoder tensors and sets stt.whisper.decoder_only=true.

The full GGUFs are unmodified copies of handy-computer/whisper-small-gguf at commit 4afdc794, converted by handy-computer for transcribe.cpp (see NOTICE). The .bin files are unmodified copies of ggerganov/whisper.cpp at commit 5359861c, under OpenAI's MIT license for Whisper (LICENSE-MIT).

Artifact	SHA-256
Encoder ZIP	`8a7eff95acc7a237731d778d2269ea7aa7338cd1190fd6ab4266be506886a625`
`whisper-small-Q8_0-decoder.gguf`	`5b8c6b995f6a6996dfe58defbfea799ac4efb0e764e1ca926007a3eeac15f359`
`whisper-small-Q8_0.gguf`	`9b9c8811bbcc82a7766f0fb0925614bdacb0923b2cc630daeac17108b655b860`
`whisper-small-Q5_K_M.gguf`	`326cd00c3e7217c751667c7c1600eaf7e0de174e186ca2c16b4bf590251c3c3b`
`whisper-small-F16.gguf`	`bef65e1ac9d012269453243243aac0d0f67792693ba6c99b584a124e0ee326fc`
`ggml-small-q5_1.bin`	`ae85e4a935d7a567bd102fe55afc16bb595bdb618e11b2fc7591bc08120411bb`
`ggml-small-q8_0.bin`	`49c8fb02b65e6049d5fa6c04f81f53b867b5ec9540406812c643f177317f779f`
`ggml-small.bin`	`1be3a9b2063867b937e64e2ec7483364a79917e157fa98c5d94b5c1fffea987b`

Runtime requirements

Extract the ZIP next to the model file, keeping the whisper-small-encoder.mlmodelc directory name. Glimpse-Speech looks for the companion beside the model. A full model file falls back to its own encoder when the companion isn't there; a decoder-only GGUF has no encoder to fall back to, so a session fails without the companion. Decoder-only GGUFs need a transcribe.cpp build with decoder-only Whisper support.

The encoder follows Apple's Neural Engine transformer layout and computes in FP16. It takes one 30-second Whisper window. Core ML may run some operations on the CPU; this is not a guarantee of exclusive Neural Engine execution.

Glimpse

This encoder speeds up on-device dictation on Apple Silicon in Glimpse, a free, open-source dictation app for Mac and Windows. The source is on GitHub.

License