[wip] update v2.1 readme (PaddlePaddle#736)

Author: lyuwenyu

CITATION.cff

@@ -0,0 +1,15 @@
+# This CITATION.cff file was generated with cffinit.
+# Visit https://bit.ly/cffinit to generate yours today!
+
+cff-version: 1.2.0
+title: Paddle Multimodal Integration and eXploration
+message: >-
+  If you use this repository, please cite it using the metadata from this file.
+type: software
+authors:
+  - given-names: PaddleMIX Authors
+repository-code: 'https://github.com/PaddlePaddle/PaddleMIX'
+repository: 'https://github.com/PaddlePaddle/PaddleMIX'
+keywords:
+  - paddlemix
+license: Apache-2.0

README.md

@@ -24,7 +24,7 @@ PaddleMIX是基于飞桨的多模态大模型开发套件，聚合图像、文
 **🔥2024.10.11 发布PaddleMIX v2.1**
 * 支持[PaddleNLP 3.0 beta](https://github.com/PaddlePaddle/PaddleNLP/releases/tag/v3.0.0-beta0)版本，抢先体验其最新功能。
 * 新增[Qwen2-VL](./paddlemix/examples/qwen2_vl/)、[InternVL2](./paddlemix/examples/internvl2/)、[Stable Diffusion 3 (SD3)](https://github.com/PaddlePaddle/PaddleMIX/blob/develop/ppdiffusers/examples/dreambooth/README_sd3.md)等前沿模型。
-* 发布自研多模数据能力标签模型[PP-InsCapTagger](./paddlemix/datacopilot/example/pp_inscaptagger/)；可用于数据的分析和过滤，试验案例表明在保持模型效果的条件下可减少50%的数据量，大幅提高训练效率。
+* DataCopilot发布自研多模数据能力标签模型[PP-InsCapTagger](./paddlemix/datacopilot/example/pp_inscaptagger/)；可用于数据的分析和过滤，试验案例表明在保持模型效果的条件下可减少50%的数据量，大幅提高训练效率。
 * 多模态大模型InternVL2、LLaVA、SD3、SDXL适配昇腾910B，提供国产计算芯片上的训推能力。
 
 **2024.09.11 更新**
@@ -145,6 +145,7 @@ sh build_env.sh
             <li><a href="paddlemix/examples/evaclip">EVA-CLIP</a></li>
             <li><a href="paddlemix/examples/llava">LLaVA</a></li>
             <li><a href="paddlemix/examples/llava">LLaVA-1.5</a></li>
+            <li><a href="paddlemix/examples/llava">LLaVA-1.6</a></li>
             <li><a href="paddlemix/examples/llava">LLaVA-NeXT</a></li>
             <li><a href="paddlemix/examples/qwen_vl">Qwen-VL</a></li>
             <li><a href="paddlemix/examples/qwen2_vl">Qwen2-VL</a></li>
@@ -169,13 +170,19 @@ sh build_env.sh
         <ul>
             <li><a href="paddlemix/examples/imagebind">ImageBind</a></li>
       </ul>
+      </ul>
+        <li><b>数据分析</b></li>
+      <ul>
+          <li><a href="./paddlemix/datacopilot/example/pp_inscaptagger/">PP-InsCapTagger</a></li>
+      </ul>
       </td>
       <td>
         <ul>
         </ul>
           <li><b>文生图</b></li>
         <ul>
            <li><a href="ppdiffusers/examples/stable_diffusion">Stable Diffusion</a></li>
+           <li><a href="ppdiffusers/examples/dreambooth/README_sd3.md">Stable Diffusion 3 (SD3)</a></li>
             <li><a href="ppdiffusers/examples/controlnet">ControlNet</a></li>
             <li><a href="ppdiffusers/examples/t2i-adapter">T2I-Adapter</a></li>
             <li><a href="ppdiffusers/examples/text_to_image_laion400m">LDM</a></li>

paddlemix/datacopilot/example/pp_inscaptagger/readme.md

@@ -110,3 +110,32 @@ LLaVA v1.5模型SFT阶段训练时，使用的指令微调数据集为[LLaVA-Ins
 | llava-1.5-7b <br> (tag 50%/our) | 70.24 | 57.12 | 78.32 | 62.14 | 37.11 | 1476 <br> 338 |
 
 通过PP-InsCapTagger的打标和优化，50%数据集与原始数据集的训练效果基本持平，大大提高了模型训练效率。
+
+
+
+## 引用
+如果在你的工作中用到`PP-InsCapTagger`，请按照下面的方式引用：
+
+<details>
+<summary> bibtex </summary>
+
+```bibtex
+
+@software{PaddleMIX_Authors_Paddle_Multimodal_Integration,
+author = {PaddleMIX Authors},
+license = {Apache-2.0},
+title = {{Paddle Multimodal Integration and eXploration}},
+url = {https://github.com/PaddlePaddle/PaddleMIX}
+}
+
+@software{Lv_Instance_Capability_Tagger_2024,
+author = {Lv, Wenyu and Huang, Kui and Zhao, Yian},
+license = {Apache-2.0},
+month = oct,
+title = {{Instance Capability Tagger: Enhancing Multimodal Data Efficiency for Model Training}},
+url = {https://github.com/lyuwenyu/PP-InsCapTagger},
+version = {1.0},
+year = {2024}
+}
+```
+</details>

paddlemix/datacopilot/nn/inscaptagger.py

@@ -39,7 +39,7 @@ def __init__(self, model_name_or_path, max_new_tokens = 4096, dtype='float16') -
 
 
     def init_model(self, model_name_or_path, max_new_tokens, dtype):
-        tokenizer = AutoTokenizerMIX.from_pretrained(model_name_or_path, use_fast=False)
+        tokenizer = AutoTokenizerMIX.from_pretrained(model_name_or_path)
         model_config = AutoConfigMIX.from_pretrained(model_name_or_path)
         model = AutoModelMIX.from_pretrained(model_name_or_path, dtype=dtype)
         model.eval()