Glimpse-Dictation/Qwen3-ASR-0.6B-coreml

Qwen3-ASR 0.6B: Core ML encoder and GGUF

This repository contains the optional Apple Silicon encoder for Qwen3-ASR 0.6B (Q8_0) in Glimpse. It is a companion to the Q8_0 GGUF, not a replacement for it: the GGUF still provides the decoder, tokenizer and model metadata, and this package moves the audio encoder onto Core ML.

File	Download size	Purpose
Qwen3-ASR-0.6B-Q8_0-encoder.mlmodelc.zip	340.3 MB	Compiled Core ML encoder
Qwen3-ASR-0.6B-Q8_0.gguf	850.4 MB	Full Q8_0 model
Qwen3-ASR-0.6B-Q8_0-decoder.gguf	639.6 MB	Decoder, tokenizer and metadata, used with the Core ML encoder

On Windows, Intel Macs, or anywhere without Core ML, use the GGUF on its own. The Core ML package requires Apple Silicon.

Provenance

The model originates from Qwen3-ASR 0.6B, licensed Apache 2.0. Qwen3-ASR-0.6B-Q8_0.gguf is an unmodified copy of handy-computer/Qwen3-ASR-0.6B-gguf at commit 7e7c69a5, converted by handy-computer for transcribe.cpp (see NOTICE). The encoder was converted from it with transcribe.cpp's scripts/convert-qwen3-asr-gguf-to-coreml.py, then zipped with ditto -c -k --keepParent.

Artifact	SHA-256
Source GGUF	f081b2d5e23bd669d92cc331d722a8a0681943b8e6f34b48996fd5c319b5acd8
Encoder ZIP	cfd2e37a30d0da1b685da0f96e82f05ef6fc208136c85efa5368ddd2ee18ad5d

Runtime requirements

Extract the ZIP next to the GGUF, keeping the Qwen3-ASR-0.6B-Q8_0-encoder.mlmodelc directory name. Glimpse-Speech looks for the companion beside the GGUF and falls back to the GGUF's own encoder when it isn't there.

The encoder holds up to 15 seconds of audio. Glimpse splits longer recordings into chunks.

Qwen3-ASR is an experimental model in Glimpse. 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 Qwen's, released under the Apache 2.0 license. This repository only converts the original checkpoint. It is not an official Qwen release. See the original model card for training data and intended use.