---
title: Leanstral-2603-8bit
canonical_url: "https://www.modelscope.cn/models/mlx-community/Leanstral-2603-8bit"
md_url: "https://www.modelscope.cn/models/mlx-community/Leanstral-2603-8bit.md"
repository: mlx-community/Leanstral-2603-8bit
last_updated: 2026-03-23
license: apache-2.0
model_type:
  - mistral3
base_model:
  - mistralai/Leanstral-2603
base_model_relation: quantized
parameters: 33.9B
tensor_type:
  - BF16
  - U32
library_name:
  - mlx
  - safetensors
  - pytorch
frameworks:
  - pytorch
downloads: 9
stars: 0
tags:
  - mlx
  - safetensors
  - mistral3
  - mistral4
  - 8-bit
  - leanstral
  - lean4
---

# Leanstral-2603-8bit

> Leanstral-2603-8bit - mlx-community 在 ModelScope 开源的模型。Leanstral-2603 MLX 8-bit

mlx-community/Leanstral-2603-8bit 是 ModelScope 魔搭社区上的 33.9B 参数机器学习模型，采用 apache-2.0 许可，基于 mistralai/Leanstral-2603 构建。

- **Repository**: mlx-community/Leanstral-2603-8bit
- **License**: apache-2.0
- **Parameters**: 33.9B
- **Base model**: mistralai/Leanstral-2603
- **Tags**: mlx, safetensors, mistral3, mistral4, 8-bit, leanstral, lean4
- **Downloads**: 9
- **Stars**: 0
- **Last updated**: 2026-03-23

Source: https://www.modelscope.cn/models/mlx-community/Leanstral-2603-8bit

---

# Leanstral-2603 MLX 8-bit

MLX 8-bit quantized version of [mistralai/Leanstral-2603](https://huggingface.co/mistralai/Leanstral-2603).

- **Architecture**: Mistral-Small-4 (119B params, 6.5B active per token, MoE with 128 experts)
- **Specialization**: Lean 4 proof engineering
- **Quantization**: 8-bit (~8.5 bits/weight), 119 GB
- **Format**: MLX safetensors (for Apple Silicon)
- **RAM required**: ~128 GB+ unified memory

## Usage

```python
from mlx_vlm import load, generate

model, processor = load("mlx-community/Leanstral-2603-8bit")
output = generate(model, processor, "Prove that the sum of two even numbers is even in Lean 4", max_tokens=4096)
print(output)
```