README.md

---
license: apache-2.0
base_model: mistralai/Codestral-22B-v0.1
library_name: mlx
pipeline_tag: text-generation
language:
- en
tags:
- codestral
- mlx
- code
- mistral
- apple-silicon
- FIM
- Fill-in-the-Middle
- code-generation
- 4bit
model-index:
  - name: Codestral-22B-Yui-MLX
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          type: code
          name: Code
        metrics:
          - type: vram
            value: 12.5 GB
---

# 🇬🇧 English | 🇨🇳 中文
> 👇 Scroll down for **Chinese version** / 向下滚动查看**中文版本**

---

## 🇬🇧 English
# Codestral-22B-Yui-MLX
This model is **CyberYui's custom-converted MLX format port** of Mistral AI's official [`mistralai/Codestral-22B-v0.1`](https://huggingface.co/mistralai/Codestral-22B-v0.1) model. No modifications, alterations, or fine-tuning of any kind were applied to the original model's weights, architecture, or parameters; this is strictly a format conversion for MLX, optimized exclusively for Apple Silicon (M1/M2/M3/M4) chips.

### 📌 Model Details
- **Base Model**: [`mistralai/Codestral-22B-v0.1`](https://huggingface.co/mistralai/Codestral-22B-v0.1)
- **Conversion Tool**: `mlx-lm 0.29.1`
- **Quantization**: 4-bit (≈12.5GB total size)
- **Framework**: MLX (native Apple GPU acceleration)
- **Use Cases**: Code completion, code generation, programming assistance, FIM (Fill-In-the-Middle)

### 🚀 How to Use
#### 1. Command Line (mlx-lm)
First, install the required package:
```bash
pip install mlx-lm
```
Then run the model directly:
```bash
mlx_lm.generate --model CyberYui/Codestral-22B-Yui-MLX --prompt "def quicksort(arr):"
```

#### 2. Python Code
```python
from mlx_lm import load, generate

# Load this model
model, tokenizer = load("CyberYui/Codestral-22B-Yui-MLX")

# Define your prompt
prompt = "Write a Python function for quicksort with comments"

# Apply chat template if available
if tokenizer.chat_template is not None:
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
messages, add_generation_prompt=True, return_dict=False
)

# Generate response
response = generate(model, tokenizer, prompt=prompt, verbose=True)
```

#### 3. LM Studio
1. Open LM Studio and log in to your Hugging Face account
2. Go to the **Publish** tab
3. Search for this model`CyberYui/Codestral-22B-Yui-MLX`
4. Download and load the model to enjoy native MLX acceleration!

### 📄 License
This model is distributed under the **Apache License 2.0**, strictly following the original model's open-source license.

---

## 🇨🇳 中文
# Codestral-22B-Yui-MLX
本模型名为 **Codestral-22B** ，是基于 **Mistral AI 官方 [`mistralai/Codestral-22B-v0.1`](https://huggingface.co/mistralai/Codestral-22B-v0.1) 模型，无任何修改并由**CyberYui**个人转换的 MLX 格式专属版本模型**，专为 **Apple Silicon（M1/M2/M3/M4）** 系列芯片深度适配。

### 📌 模型详情
- **基础模型**：[`mistralai/Codestral-22B-v0.1`](https://huggingface.co/mistralai/Codestral-22B-v0.1)
- **转换工具**：`mlx-lm 0.29.1`
- **量化精度**：4-bit（总大小约 12.5GB）
- **运行框架**：MLX（原生苹果 GPU 加速）
- **适用场景**：代码补全、代码生成、编程辅助、FIM（中间填充）

### 🚀 使用方法
#### 1. 命令行（mlx-lm）
首先安装依赖包：
```bash
pip install mlx-lm
```
然后直接运行模型：
```bash
mlx_lm.generate --model CyberYui/Codestral-22B-Yui-MLX --prompt "def quicksort(arr):"
```

#### 2. Python 代码
```python
from mlx_lm import load, generate

# 加载本模型
model, tokenizer = load("CyberYui/Codestral-22B-Yui-MLX")

# 定义你的提示词
prompt = "写一个带注释的Python快速排序函数"

# 应用对话模板
if tokenizer.chat_template is not None:
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
messages, add_generation_prompt=True, return_dict=False
)

# 模型会生成回复
response = generate(model, tokenizer, prompt=prompt, verbose=True)
```

#### 3. LM Studio 使用
1. 打开 LM Studio，登录你的 Hugging Face 账号
2. 进入 **Publish** 标签页
3. 搜索 `CyberYui/Codestral-22B-Yui-MLX`
4. 下载并加载模型，即可享受原生 MLX 版本的 Codestral 模型了！

### 📄 开源协议
本模型遵循 **Apache License 2.0** 协议分发，严格遵守原模型的开源要求。
```