Downloads · 30 days
0
qwen3-4b-coreai
geekaholic/qwen3-4b-coreai
qwen3-4b-coreai is a text generation model from geekaholic. Use it when you need the model to write or continue text. The card lists the license as apache-2.0.
An Apple Core AI (.aimodel) export of Qwen/Qwen3-4B, packaged for on-device inference in Omni Chat on macOS 27 and iPadOS 27.
Access
Public
Updated Sep 21, 2026
Repo size
12.2 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.gz10 GB · 100%
At a glance
- Task
- Text Generation
- License
- apache-2.0
- Access
- Public
- Created
- Sep 20, 2026
- Updated
- Sep 21, 2026
- SHA
- c821c6d5
Try a prompt
Base models
Model card
From the Hugging Face model README
Qwen3 4B — Core AI export
An Apple Core AI (.aimodel) export of Qwen/Qwen3-4B,
packaged for on-device inference in Omni Chat on
macOS 27 and iPadOS 27.
This repository hosts the release artifact only. It is downloaded and verified by the Omni Chat
Go core against a checksum manifest embedded in the application
(core/internal/localmodel/artifacts.json) — every file in the archive is hashed individually,
extraction is bounded, and installation is atomic.
Provenance
| Source model | Qwen/Qwen3-4B at revision 1cfa9a7208912126459214e8b04321603b3df60c |
| Export recipe | apple/coreai-models at revision 3f109efd54273391f9fd9f5f5b3d8c6e99836d55 |
| Context window | 8,192 tokens |
| Max output | 4,096–6,144 tokens, by device memory |
Two exports, one per platform, both using their own uniform 4-bit preset (macOS: 4bit; iOS:
4bit_weight_palettized_group32, its documented default). Both compile the model for the specific
device on first load; the export itself is not ahead-of-time compiled -- see
docs/coreai.md
for why that isn't shipped for iPad.
| macOS | iPadOS | |
|---|---|---|
| Archive | qwen3-4b-macos.tar.gz | qwen3-4b-ios.tar.gz |
| Size | 1,980,006,225 bytes | 1,964,191,754 bytes |
| SHA-256 | e603a73bb715c91d2dec10460ba87f50500683fa06dc2aad17f2760f3796d050 | 66d793a0b11169d4d0c99409408e963739b218b40b02c1f063e1bf178bae16ba |
The archive carries its own tokenizer, so inference needs no Hugging Face access. It also carries
PROVENANCE.json, LICENSE (Apache-2.0, from the source model) and COREAI-LICENSE.
Requirements
macOS 27 or iPadOS 27 on Apple Silicon with at least 8 GB of memory. Text, thinking and tool calls are supported; there is no vision or audio path in this export.
Reproducing
See docs/coreai.md in the
Omni Chat repository. The export is deterministic from the two pinned revisions above.
- Task
- Text Generation
- License
- apache-2.0
- Created
- Sep 20, 2026
- Updated
- Sep 21, 2026