# LLaVA-NeXT: Tackling Multi-image, Video, and 3D in Large Multimodal Models

## 论文

`LLaVA-NeXT-Interleave: Tackling Multi-image, Video, and 3D
 in Large Multimodal Models`

* https://arxiv.org/pdf/2407.07895

## 模型结构

参考[README.md](../README.md)

## 算法原理

在[README.md](../README.md)基础上，LLaVA-NeXT-Interleave的核心是通过统一的数据格式和联合训练策略，实现多模态任务的泛化与迁移。

## 数据集

无

## 训练

无

## 推理

在 inference分支中

### 原生

```bash
python ../playground/demo/interleave_demo.py --model_path path/to/ckpt
```

### hf

```bash
python inference_hf.py
```

注意：运行前需要修改脚本中相应路径。

## result

![alt text](readme_imgs/result.png)

### 精度

无

## 应用场景

参考[README.md](../README.md)

## 预训练权重

|model|url|
|:---:|:---:|
|llava-next-interleave-qwen-7b|[hf](https://huggingface.co/lmms-lab/llava-next-interleave-qwen-7b) \| [SCNet](http://113.200.138.88:18080/aimodels/lmms-lab/llava-next-interleave-qwen-7b) |
|llava-next-interleave-qwen-0.5b|[hf](https://hf-mirror.com/lmms-lab/llava-next-interleave-qwen-0.5b) \| [SCNet](http://113.200.138.88:18080/aimodels/lmms-lab/llava-next-interleave-qwen-0.5b.git) |
|llava-interleave-qwen-0.5b-hf|[hf](https://huggingface.co/llava-hf/llava-interleave-qwen-0.5b-hf) \| [SCNet](http://113.200.138.88:18080/aimodels/llava-hf/llava-interleave-qwen-0.5b-hf) |
|llava-interleave-qwen-7b-hf|[hf](https://huggingface.co/llava-hf/llava-interleave-qwen-7b-hf) \| [SCNet](http://113.200.138.88:18080/aimodels/llava-hf/llava-interleave-qwen-7b-hf) |
|llava-interleave-qwen-7b-dpo-hf|[hf](https://huggingface.co/llava-hf/llava-interleave-qwen-7b-dpo-hf) \| [SCNet](http://113.200.138.88:18080/aimodels/llava-hf/llava-interleave-qwen-7b-dpo-hf) |


模型下载后保存至`ckpts`（需自行创建）.

## 源码仓库及问题反馈

参考[README.md](../README.md)

## 参考资料

* https://github.com/LLaVA-VL/LLaVA-NeXT/blob/main/docs/LLaVA-NeXT-Interleave.md
* https://llava-vl.github.io/blog/2024-06-16-llava-next-interleave/