---
title: YOLO26s-OptiQ-6bit
canonical_url: "https://www.modelscope.cn/models/mlx-community/YOLO26s-OptiQ-6bit"
md_url: "https://www.modelscope.cn/models/mlx-community/YOLO26s-OptiQ-6bit.md"
repository: mlx-community/YOLO26s-OptiQ-6bit
last_updated: 2026-09-28
license: agpl-3.0
base_model:
  - Ultralytics/YOLO26
base_model_relation: finetune
parameters: 2.3M
tensor_type:
  - F32
  - U32
library_name:
  - mlx
  - safetensors
  - pytorch
frameworks:
  - pytorch
downloads: 8
stars: 0
tags:
  - mlx
  - quantized
  - mixed-precision
  - yolo
  - yolo26
  - object-detection
  - optiq
  - apple-silicon
---

# YOLO26s-OptiQ-6bit

> YOLO26s-OptiQ-6bit - mlx-community 在 ModelScope 开源的模型。Mixed-precision quantized YOLO26s for Apple Silicon via optiq

mlx-community/YOLO26s-OptiQ-6bit 是 ModelScope 魔搭社区上的 2.3M 参数机器学习模型，采用 agpl-3.0 许可，基于 Ultralytics/YOLO26 构建。

- **Repository**: mlx-community/YOLO26s-OptiQ-6bit
- **License**: agpl-3.0
- **Parameters**: 2.3M
- **Base model**: Ultralytics/YOLO26
- **Tags**: mlx, quantized, mixed-precision, yolo, yolo26, object-detection, optiq, apple-silicon
- **Downloads**: 8
- **Stars**: 0
- **Last updated**: 2026-09-28

Source: https://www.modelscope.cn/models/mlx-community/YOLO26s-OptiQ-6bit

---

# YOLO26s-OptiQ-6bit

> Mixed-precision quantized YOLO26s for Apple Silicon via optiq

This is a mixed-precision quantized version of [YOLO26s](https://github.com/ultralytics/ultralytics) in MLX format, optimized with [mlx-optiq](https://mlx-optiq.com) for Apple Silicon inference via [yolo-mlx](https://pypi.org/project/yolo-mlx/).

## Quantization Details

| Property | Value |
|---|---|
| Target BPW | 6.0 |
| Achieved BPW | 5.97 |
| Layers at 4-bit | 11 |
| Layers at 8-bit | 115 |
| Original size | 38.4 MB |
| Quantized size | 8.9 MB |
| Compression | 4.3x |

## Benchmark Results (COCO128)

| Model | Total Detections | Avg/Image |
|---|---|---|
| **optiq 6-bit** | **633** | **4.9** |
| Original (FP32) | 681 | 5.3 |

Detection delta: -48 (-7.0%) at 4.3x compression.

## Usage

Requires `mlx-optiq` and `yolo-mlx`:

```bash
pip install mlx-optiq yolo-mlx
```

```python
from optiq.models.yolo import load_quantized_yolo

model = load_quantized_yolo("mlx-community/YOLO26s-OptiQ-6bit")
results = model.predict("image.jpg")
```

## How optiq Works

optiq measures each conv layer's sensitivity via KL divergence on detection outputs, then assigns optimal per-layer bit-widths using greedy knapsack optimization. Sensitive layers (detection head, feature pyramid) get 8-bit precision while robust backbone layers get 4-bit.



## Article

For more details on the methodology and results, see: [Not All Layers Are Equal](https://x.com/latent_node/status/2028412948167942334?s=20)

## Credits

- **Quantization:** [mlx-optiq](https://mlx-optiq.com)
- **Base model:** [YOLO26](https://github.com/ultralytics/ultralytics) by Ultralytics
- **MLX runtime:** [yolo-mlx](https://pypi.org/project/yolo-mlx/)
- **Framework:** [MLX](https://github.com/ml-explore/mlx) by Apple