Qwen3-8B for Core AI on iOS
Core AI export of Qwen/Qwen3-8B, developed by the Qwen team.
The upstream revision used for this export is b968826d9c46dd6066d109eabc6255188de91218.
The accompanying LICENSE is copied from that revision without modification.
Exported with Apple coreai-models
at e7b24da85ea64a77d26324d7ce9607de9b955f57, using preset qwen3_8b_mixed_4bit_8bit for iOS
with an 8,192-token context. EXPORT.md records the export command. The export changes
representation and quantization; it retains the upstream tokenizer and chat template.
The folder contains the .aimodel artifact, export metadata and tokenizer files.
It requires Core AI on iOS 27 or later. It is not a Transformers or MLX weights folder.
Runtime performance and the supported device tiers remain subject to device verification;
exported file sizes do not establish a device's runtime memory requirement.
Files and attribution
Model weights and tokenizer: Qwen team, Apache License 2.0. Conversion tools:
Apple coreai-models. This repository distributes
the export for Hearth, a local writing and document assistant. See EXPORT.md for
reproduction details and the upstream model card for training, evaluation and limitations.