Glimpse-Dictation/Whisper-Base-coreml

Whisper Base: Core ML encoder and full models

This repository contains the optional Apple Silicon encoder for Whisper Base in Glimpse. It is a companion to a Whisper GGUF or whisper.cpp ggml model, not a replacement for it: the model file still provides the decoder, tokenizer and metadata, and this package moves the audio encoder onto Core ML. Glimpse uses it for every Whisper Base quantization.

File	Download size	Purpose
whisper-base-encoder.mlmodelc.zip	37.9 MB	Compiled Core ML encoder

On Windows, Intel Macs, or anywhere without Core ML, use the model file on its own. 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-base-Q8_0.gguf	85.0 MB	Full GGUF, Q8_0
whisper-base-Q5_K_M.gguf	63.8 MB	Full GGUF, Q5_K_M
whisper-base-F16.gguf	151.1 MB	Full GGUF, F16
ggml-base-q5_1.bin	59.7 MB	whisper.cpp ggml, Q5_1
ggml-base-q8_0.bin	81.8 MB	whisper.cpp ggml, Q8_0
ggml-base.bin	148.0 MB	whisper.cpp ggml, F16

Provenance

The model originates from OpenAI Whisper Base, licensed Apache 2.0. The encoder was converted from whisper.cpp's F16 ggml-base.bin with scripts/convert-whisper-gguf-to-coreml.py from our transcribe.cpp fork, then zipped with ditto -c -k --keepParent.

The full GGUFs are unmodified copies of handy-computer/whisper-base-gguf at commit 42fcd207, 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	f38ea79465a06476d59a7e60bba129df6fa0823264f22242c093b604f4c3e533
whisper-base-Q8_0.gguf	81c069428bc8a24551a8169cf31cf09bcfd9d4cf50389ae281323c9aa9648c81
whisper-base-Q5_K_M.gguf	8e0feb7bc35780353cf31821018e601bb7b7cff6c9a0e17ada5a5db23f4db867
whisper-base-F16.gguf	38ab6b0ed742e9eded4d5a2ba7fc34d44fc28cdb71d6360f6321d4b306ef8039
ggml-base-q5_1.bin	422f1ae452ade6f30a004d7e5c6a43195e4433bc370bf23fac9cc591f01a8898
ggml-base-q8_0.bin	c577b9a86e7e048a0b7eada054f4dd79a56bbfa911fbdacf900ac5b567cbb7d9
ggml-base.bin	60ed5bc3dd14eea856493d334349b405782ddcaf0028d4b5df4088345fba2efe

Runtime requirements

Extract the ZIP next to the model file, keeping the whisper-base-encoder.mlmodelc directory name. Glimpse-Speech looks for the companion beside the model and falls back to the model's own encoder when it isn't there.

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

The model and its weights are OpenAI's, released under the Apache 2.0 license (LICENSE). This repository only converts and mirrors the original checkpoint. It is not an official OpenAI release. See the original model card for training data and intended use.