Downloads · 30 days
19
5% of all-time downloads
greyscope-v2-qwen3.5-4b-mlx-4bit
yaoandy107/greyscope-v2-qwen3.5-4b-mlx-4bit
greyscope-v2-qwen3.5-4b-mlx-4bit is a text classification model from yaoandy107. Use it when you need a label for a piece of text. It is set up for mlx. The card lists the license as apache-2.0.
This is the recommended Greyscope build for Apple Silicon. It is 2.4 GB and returns the same continuous aiinvolvement score and human / AI-edited / AI-generated labels as the bf16 model.
All-time downloads
420
Public
Parameters
4.2B
2.4 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors2.4 GB · 99%
Parameter types
How the weights are stored.
U324.2B · 100%
Try a prompt
Base models
Model card
From the Hugging Face model README
Greyscope v2 MLX 4-bit
This is the recommended Greyscope build for Apple Silicon. It is 2.4 GB and returns the same
continuous ai_involvement score and human / AI-edited / AI-generated labels as the
bf16 model.
Quick start
git clone https://github.com/yaoandy107/greyscope
cd greyscope
uv sync --extra mac
uv run greyscope "Paste a paragraph here."
Mac performance
Measured on an M1 Pro with 32 GB unified memory and batch size 1:
| Build | Peak memory | Load time | 128 tokens | 512 tokens | 1,024 tokens |
|---|---|---|---|---|---|
| MLX 4-bit | 3.0 GB | 2.0 s | 0.66 s | 2.43 s | 5.13 s |
| Transformers bf16 | 9.4 GB | 16.4 s | 0.74 s | 2.54 s | 5.01 s |
Quantization check
MLX 4-bit matched bf16 quality across 3,007 external benchmark rows. See the release metrics for details.
Public evaluations, calibration, training details, and limitations are documented on the bf16 model card.
- Task
- Text Classification
- Library
- mlx
- Type
- qwen3_5_text
- License
- apache-2.0
- Languages
- en, ja, zh
- Created
- Jul 26, 2026
- Updated
- Jul 28, 2026