---
title: XYZ-Aquila-mini-OptiQ-4bit
canonical_url: "https://www.modelscope.cn/models/latentnode/XYZ-Aquila-mini-OptiQ-4bit"
md_url: "https://www.modelscope.cn/models/latentnode/XYZ-Aquila-mini-OptiQ-4bit.md"
repository: latentnode/XYZ-Aquila-mini-OptiQ-4bit
chinese_name: XYZ-Aquila-mini-OptiQ-4bit
last_updated: 2026-07-27
license: apache-2.0
pipeline_tag: text-generation
tasks:
  - text-generation
model_type:
  - qwen3_5_moe
architectures:
  - Qwen3_5MoeForConditionalGeneration
base_model:
  - XYZAILab/XYZ-Aquila-mini
base_model_relation: quantized
parameters: 10.1B
tensor_type:
  - U32
  - BF16
library_name:
  - mlx
  - safetensors
downloads: 14
stars: 0
tags:
  - mlx
  - optiq
  - mixed-precision
  - qwen3.6
  - agentic-search
---

# XYZ-Aquila-mini-OptiQ-4bit

> XYZ-Aquila-mini-OptiQ-4bit - latentnode 在 ModelScope 开源的模型。XYZ-Aquila-mini-OptiQ-4bit

latentnode/XYZ-Aquila-mini-OptiQ-4bit 是 ModelScope 魔搭社区上的 10.1B 参数text-generation模型，采用 apache-2.0 许可，基于 XYZAILab/XYZ-Aquila-mini 构建。

- **Repository**: latentnode/XYZ-Aquila-mini-OptiQ-4bit
- **License**: apache-2.0
- **Tasks**: text-generation
- **Parameters**: 10.1B
- **Base model**: XYZAILab/XYZ-Aquila-mini
- **Tags**: mlx, optiq, mixed-precision, qwen3.6, agentic-search
- **Downloads**: 14
- **Stars**: 0
- **Last updated**: 2026-07-27

Source: https://www.modelscope.cn/models/latentnode/XYZ-Aquila-mini-OptiQ-4bit

---

# XYZ-Aquila-mini-OptiQ-4bit

An [OptiQ](https://mlx-optiq.com) mixed-precision quantization of
[XYZAILab/XYZ-Aquila-mini](https://huggingface.co/XYZAILab/XYZ-Aquila-mini),
built for Apple Silicon with MLX.

OptiQ measures each layer's sensitivity and assigns a per-layer bit-width
(4-bit or 8-bit here) instead of quantizing every layer the same, so the layers
that need the precision keep it.

- **Method:** OptiQ data-driven mixed precision (`optiq_mixed_precision`)
- **Average:** ~5.0 bits per weight (94 layers at 8-bit, 296 at 4-bit)
- **Architecture:** `qwen3_5_moe` sparse MoE
- **Reasoning:** append `:think` / `:no-think` to the model id to control the thinking channel

## Use

```bash
pip install mlx-optiq
optiq serve --model mlx-community/XYZ-Aquila-mini-OptiQ-4bit
```

This is a large mixture-of-experts model. `optiq serve` streams the experts off
SSD automatically, so it runs with a small resident footprint (about 4 GB here)
rather than loading all weights at once. See [mlx-optiq.com](https://mlx-optiq.com)
for the CLI, the local Lab UI, and the OptiQ Code agent.