PM4Bench-Release commited on
Commit
ce986fe
·
1 Parent(s): 3614893

Update paper links, resource navigation, and citation

Browse files
Files changed (1) hide show
  1. README.md +38 -1
README.md CHANGED
@@ -13,9 +13,32 @@ datasets:
13
  - DatasetMan/PM4Bench-QGO-Train
14
  ---
15
 
 
 
16
  # QGO-8B
17
 
18
- QGO-8B is an OCR-centric GRPO checkpoint derived from
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
19
  [`Qwen/Qwen3-VL-8B-Thinking`](https://huggingface.co/Qwen/Qwen3-VL-8B-Thinking).
20
  This repository contains the complete global-step-200 BF16 checkpoint used in
21
  the PM4Bench paper.
@@ -76,3 +99,17 @@ QGO-8B targets multilingual OCR robustness. It inherits limitations and risks
76
  from the Qwen base model and is not guaranteed to improve every downstream
77
  task or language. Coordinate outputs, OCR transcriptions, and long-form
78
  reasoning should be validated before use in consequential applications.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
13
  - DatasetMan/PM4Bench-QGO-Train
14
  ---
15
 
16
+ <div align="center">
17
+
18
  # QGO-8B
19
 
20
+ **OCR-centric reinforcement learning for multilingual LVLMs**
21
+
22
+ [![arXiv](https://img.shields.io/badge/arXiv-2503.18484v3-B31B1B.svg?logo=arxiv&logoColor=white)](https://arxiv.org/abs/2503.18484v3)
23
+ [![Project Page](https://img.shields.io/badge/Project-Page-145C73.svg?logo=githubpages&logoColor=white)](https://little-bird-vodka.github.io/PM4Bench/)
24
+ [![Code](https://img.shields.io/badge/GitHub-Code-181717.svg?logo=github&logoColor=white)](https://github.com/opendatalab/PM4Bench)
25
+ [![Benchmark](https://img.shields.io/badge/Hugging_Face-Benchmark-FFD21E.svg)](https://huggingface.co/datasets/songjhPKU/PM4Bench)
26
+ [![Training Data](https://img.shields.io/badge/Hugging_Face-QGO_Train-FFD21E.svg)](https://huggingface.co/datasets/DatasetMan/PM4Bench-QGO-Train)
27
+
28
+ </div>
29
+
30
+ ## Overview
31
+
32
+ The paper [*Benchmarking and Boosting Multilingual Capabilities of LVLMs via
33
+ OCR-Centric Reinforcement Learning*](https://arxiv.org/abs/2503.18484v3)
34
+ introduces the strictly parallel PM4Bench benchmark and finds that OCR is a key
35
+ source of cross-lingual performance gaps under vision-only input. QGO trains
36
+ on fully synthesized OCR data with a dual reward for recognition accuracy and
37
+ reasoning-preserving response structure. It improves average score on every
38
+ PM4Bench task and setting and transfers gains to external multilingual OCR
39
+ benchmarks.
40
+
41
+ QGO-8B is the OCR-centric GRPO checkpoint derived from
42
  [`Qwen/Qwen3-VL-8B-Thinking`](https://huggingface.co/Qwen/Qwen3-VL-8B-Thinking).
43
  This repository contains the complete global-step-200 BF16 checkpoint used in
44
  the PM4Bench paper.
 
99
  from the Qwen base model and is not guaranteed to improve every downstream
100
  task or language. Coordinate outputs, OCR transcriptions, and long-form
101
  reasoning should be validated before use in consequential applications.
102
+
103
+ ## Cite Us
104
+
105
+ ```bibtex
106
+ @misc{gao2026benchmarkingboostingmultilingualcapabilities,
107
+ title={Benchmarking and Boosting Multilingual Capabilities of LVLMs via OCR-Centric Reinforcement Learning},
108
+ author={Junyuan Gao and Jiahe Song and Jiang Wu and Runchuan Zhu and Guanlin Shen and Shasha Wang and Xingjian Wei and Haote Yang and Weijia Li and Bin Wang and Lijun Wu and Conghui He},
109
+ year={2026},
110
+ eprint={2503.18484},
111
+ archivePrefix={arXiv},
112
+ primaryClass={cs.CV},
113
+ url={https://arxiv.org/abs/2503.18484},
114
+ }
115
+ ```