Downloads · 30 days
46
17% of all-time downloads
Geer-Qwen3.8-27B-4-8bit-MLX
ilyakam/Geer-Qwen3.8-27B-4-8bit-MLX
Geer-Qwen3.8-27B-4-8bit-MLX is a image-text-to-text model from ilyakam. Use it for the image-text-to-text task on the model card, and read the license before you ship it in a product. It is set up for mlx. The card lists the license as apache-2.0.
This is Geer's independently produced mixed 4/8-bit MLX conversion of Qwen3.8-27B for 48 GB Apple-Silicon Macs. Geer does not offer or claim a 32 GB Qwen profile; those machines continue to use Ornith. This is not a n…
All-time downloads
277
Public
Parameters
27.8B
24.2 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors24.2 GB · 100%
Parameter types
How the weights are stored.
U3224.4B · 88%
Try a prompt
Base models
Model card
From the Hugging Face model README
Geer Qwen 3.8 27B (4/8-bit MLX)
This is Geer's independently produced mixed 4/8-bit MLX conversion of Qwen3.8-27B for 48 GB Apple-Silicon Macs. Geer does not offer or claim a 32 GB Qwen profile; those machines continue to use Ornith. This is not a new foundation model.
Large feed-forward matrices use affine 4-bit quantization. Attention and
linear-attention projections use 8-bit quantization; the vision tower,
embeddings, output head, norms, and native MTP tensors remain unquantized. The
build uses group size 64 and upstream revision
1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0.
Upstream benchmark results describe the BF16 checkpoint, not this conversion. This artifact retained 100% of the qualified 6-bit aggregate pass rate in the initial 128 GB host canaries. Its 48 GB support profile uses a 128K context ceiling. That tier remains provisional until exercised on a real 48 GB Mac.
- Task
- Image-Text-to-Text
- Library
- mlx
- Type
- qwen3_5
- License
- apache-2.0
- Languages
- en
- Created
- Aug 15, 2026
- Updated
- Aug 15, 2026