---
title: next-trainer-portable
canonical_url: "https://www.modelscope.cn/datasets/windsing/next-trainer-portable"
md_url: "https://www.modelscope.cn/datasets/windsing/next-trainer-portable.md"
repository: windsing/next-trainer-portable
chinese_name: next-trainer-portable
last_updated: 2026-09-15
license: "Apache License 2.0"
storage_size: "22 GB"
downloads: 280
stars: 1
---

# next-trainer-portable

> next-trainer-portable - windsing 在 ModelScope 开源的数据集。next-trainer的数据集

windsing/next-trainer-portable 是 ModelScope 魔搭社区上的数据集，存储大小 22 GB，采用 Apache License 2.0 许可。

- **Repository**: windsing/next-trainer-portable
- **License**: Apache License 2.0
- **Storage size**: 22 GB
- **Downloads**: 280
- **Stars**: 1
- **Last updated**: 2026-09-15

Source: https://www.modelscope.cn/datasets/windsing/next-trainer-portable

---

# Next Trainer 整合包（Windows）

本仓库专门存放 **[Next Trainer](https://github.com/wochenlong/lora-scripts-next)**（仓库名 `lora-scripts-next`）的 **Windows 便携整合包**，方便国内高速下载。

| 渠道 | 内容 |
|------|------|
| **本仓（魔搭）** | 完整包（含 Anima Fast 运行时）等大文件 |
| [GitHub Releases](https://github.com/wochenlong/lora-scripts-next/releases) | 轻量 lite 包（&lt; 2 GB，不含 Fast） |

源码与 Issue：**https://github.com/wochenlong/lora-scripts-next**

---

## 当前版本：v2.9.2-beta.3（内测）

> **版本约定**：内测一律使用 **`2.9.x`**（如 `2.9.2-beta.3`）；**正式版才用 `3.0.0`**。  
> 本包基于 `dev` 的 Vue3 信息架构，**不替代** `main` 稳定版 v2.9.1。

### 相对 beta.2 的修复

- 冷启动依赖：`tensorboard==2.14.0` 与 `protobuf==3.20.3` 兼容，避免首次 `pip install` 硬失败

### 下载文件

| 文件 | 体积（约） | 说明 |
|------|------------|------|
| `releases/v2.9.2-beta.3/Next-Trainer-v2.9.2-beta.3-full.7z` | 2.3 GB | **完整包**：含 Anima Fast `.venv` + 内置 WD 打标模型 |
| `releases/v2.9.2-beta.3/Next-Trainer-v2.9.2-beta.3-lite.7z` | 0.39 GB | **轻量包**：不含 Fast；GitHub 亦有同款 |

页面右侧 / Files 中按路径下载即可。也可用 ModelScope SDK：

```python
from modelscope import snapshot_download
snapshot_download(
    "windsing/next-trainer-portable",
    repo_type="dataset",
    allow_patterns="releases/v2.9.2-beta.3/*",
    local_dir="./next-trainer-portable",
)
```

---

## 使用方法

1. 用 [7-Zip](https://www.7-zip.org/) 解压到**非中文、非空格**路径（建议 SSD）
2. 双击根目录 **`run_gui.bat`**
3. **首次启动**需联网安装主环境依赖（PyTorch 等，约数 GB）
4. 浏览器打开 **http://127.0.0.1:28000**
5. 反馈问题时请附上版本号 **`2.9.2-beta.3`**（设置侧栏可见）

### 完整包 vs 轻量包

| | full（本仓主推） | lite |
|--|------------------|------|
| Anima Fast 运行时 | 已预装 | 需在「设置 → 训练引擎」页内安装 |
| 默认 WD 打标模型 | 已内置 | 已内置 |
| 适合 | 想开箱用 Fast / 网盘与魔搭下载 | GitHub / 体积敏感 |

---

## 历史版本

- `releases/v2.9.2-beta.2/`（冷启动依赖有已知问题，请改用 beta.3）
- `releases/v2.9.2-beta.1/`

---

## 目录结构（解压后）

```text
<解压根>/
  run_gui.bat
  python_embeded/
  SD-Trainer/
  tagger-models/wd14/wd14-convnextv2-v2/
  tokenizer-cache/
```

包内项目目录名仍为 `SD-Trainer/`（启动契约）；产品品牌为 **Next Trainer**。

---

## 许可与致谢

- 本项目：Apache-2.0（见 GitHub 仓库）
- 第三方组件见仓库 `NOTICE.md` / `docs/credits.md`
- Anima Fast 运行时基于 [sorryhyun/anima_lora](https://github.com/sorryhyun/anima_lora)（MIT）
