NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI
PyPI, #4252 most downloaded on PyPI
Low-code development tool based on PaddlePaddle.
Latest stable
3.7.2
Last release 5 months ago
25 Jun 2026
Ships fairly regularly
a new release about every 2 weeks
Most releases are documented
notes for 38 of the last 60 stable releases
Nothing withdrawn
no release was ever pulled
7 years old
92 releases · first in 2020
Drag across the ruler to compare, or tap a release to open it.
Support HPI and HPS for PP-OCRv6
Support HPI and HPS for PP-OCRv6
update docs ( #5156 ) * update docs * fix
## 2026.6.11 v3.7.0 released - Release PP-OCRv6. ## 2026.6.11 v3.7.0 发布 - 发布 PP-OCRv6。 Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare
Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.6.1...v3.7.0
Your coding agent can read these notes before it upgrades. Set up the MCP server →
* Pipelines such as PaddleOCR-VL now support parsing multi-page TIFF files.
Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.6.0...v3.6.1
Supported the inference and deployment functionality for PaddleOCR-VL-1.6.
Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.5.2...v3.6.0
Models such as SLANet and PP-FormulaNet_plus-L now support inference with the Transformers backend.
Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.5.1...v3.5.2
The PP-OCRv5 multilingual model supports inference with the Transformers backend.
Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.5.0...v3.5.1
Supports switching the underlying inference engine, with options such as PaddlePaddle framework and Transformers.
Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.4.0...v3.5.0
Supported training and static-graph export for PP-DocLayoutV3.
Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.4.2...v3.4.3
The locally deployed services of PaddleOCR-VL and PaddleOCR-VL-1.5 have removed the default PDF page limit.
matmul_add_act_fuse_pass to exclude PP-DocLayoutV3 from this fusion logic.matmul_add_act_fuse_pass 算子融合影响的模型范围,仅在 PP-DocLayoutV3 时,不做该算子融合。Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.4.1...v3.4.2
Fixed an issue in PaddleOCR-VL where charts and stamps might not be recognized after consecutive inference runs.
genai-vllm-server plugin, added an upper bound on the supported transformers version to prevent compatibility issues during installation.genai-vllm-server 插件,新增对 transformers 版本上界的限制,防止安装时出现兼容性问题。Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.4.0...v3.4.1
Release the PaddleOCR-VL-1.5 complex document parsing solution.
Release the PaddleOCR-VL-1.5 complex document parsing solution.
PaddleOCR-VL-1.5 is a new iterative version of the PaddleOCR-VL series. Based on comprehensive optimization of the core capabilities of version 1.0, the model achieves 94.5% accuracy on the authoritative document parsing benchmark OmniDocBench v1.5, surpassing top global general-purpose large models and document parsing–specific models.
PaddleOCR-VL-1.5 innovatively supports irregular-shaped bounding box localization of document elements, enabling excellent performance in real-world application scenarios such as scanning, skew, warping, screen-photography, and complex illumination, achieving comprehensive SOTA performance. In addition, the model further integrates seal recognition and spotting tasks, with key metrics continuing to lead mainstream models.
You can use it online on the PaddleOCR official website or call the model API.
Add support for calling MLX-VLM inference services.
The PP-StructureV3 service supports the prettifyMarkdown and showFormulaNumber parameters, with functionality aligned with local inference.
Upgrade the PaddleOCR-VL concatenate-pages method to restructure-pages, supporting reorganization of multi-page results without changing the total number of pages, with more flexible usage.
Fix potential memory issues caused by the non-thread-safe PDF rendering library in multi-threaded invocation scenarios.
The parameter validation logic of production services such as general OCR has been optimized, so that in more cases of invalid input parameters, a 422 status code is returned instead of 500.
For GenAIClient, implement graceful exit of the asynchronous event loop to improve system stability and the reliability of resource release.
发布 PaddleOCR-VL-1.5 复杂文档解析方案。
PaddleOCR-VL-1.5 是 PaddleOCR-VL 系列的全新迭代版本。在全面优化 1.0 版本核心能力的基础上,该模型在文档解析权威评测集 OmniDocBench v1.5 上斩获了 94.5% 的高精度,超越了全球的顶尖通用大模型及文档解析专用模型。
PaddleOCR-VL-1.5 创新性地支持了文档元素的异形框定位,使得 PaddleOCR-VL-1.5 在扫描、倾斜、弯折、屏幕拍摄及复杂光照等真实落地场景中均表现卓越,实现了全面的 SOTA。此外,模型进一步集成了印章识别与文本检测识别任务,关键指标持续领跑主流模型。
您可以在 PaddleOCR官网 在线使用或者调用该模型的API。
新增对 MLX-VLM 推理服务的调用支持。
PP-StructureV3 服务支持 prettifyMarkdown 和 showFormulaNumber 参数,功能与本地推理对齐。
将 PaddleOCR-VL 的 concatenate-pages 方法升级为 restructure-pages,支持在不改变总页数的情况下重新整合多页结果,用法更加灵活。
修复在多线程调用场景下,PDF 渲染库因非并发安全导致的潜在内存问题。
优化了通用 OCR 等产线服务的参数校验逻辑,在更多传入无效参数的场景下返回 422 状态码,而非 500 状态码。
针对 GenAIClient,实现异步事件循环的优雅退出,提升系统稳定性与资源释放可靠性。
Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.3.13...v3.4.0
PaddleOCR-VL adds a new concatenate_pages method for concatenating multi-page parsing results, supporting preservation of multi-level heading structur
concatenate_pages method for concatenating multi-page parsing results, supporting preservation of multi-level heading structures and merging tables that span across pages.min_pixels and max_pixels parameters on each prediction.cyrillic_PP-OCRv5_mobile_rec model could not enable high-performance inference under PaddlePaddle 3.1.1 + CUDA 12.concatenate_pages 方法,用于拼接多页解析结果,支持保留多级标题结构、合并跨页表格。GenAIClient在构造时支持指定自定义模型名称。min_pixels 和 max_pixels 参数。cyrillic_PP-OCRv5_mobile_rec 模型在 PaddlePaddle 3.1.1 + CUDA 12 环境下无法启用高性能推理的问题。Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.3.12...v3.3.13
Fixed an issue where headers and footers were missing in the Markdown output of PP-StructureV3.
Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.3.11...v3.3.12
Fixed an issue where memory kept increasing during repeated inference.
markdown_ignore_labels to PP-StructureV3 and PaddleOCR-VL, which controls the element types to be ignored in Markdown output; pipeline outputs now include additional information such as the number of pages and image size.merge_layout_blocks parameter.PADDLE_PDX_DISABLE_MODEL_SOURCE_CHECK, resolving model loading lag in offline environments.markdown_ignore_labels,用于控制在 Markdown 输出中需忽略的元素类型;产线输出结果中新增文件页数、图像尺寸等信息。merge_layout_blocks 控制是否对图像块进行合并。PADDLE_PDX_DISABLE_MODEL_SOURCE_CHECK 绕过模型源检查,解决离线环境下模型加载卡顿的问题。Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.3.10...v3.3.11
Optimized the network implementation of the PaddleOCR-VL-0.9B model, significantly reducing GPU memory usage on devices with Compute Capability ≥ 8.
Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.3.9...v3.3.10
Fixed an issue where PP-DocLayoutV2 exhibited abnormal accuracy on CPU.
Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.3.8...v3.3.9
Fixed installation bugs in the vLLM and SGLang plugins.
Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.3.7...v3.3.8
PaddleOCR-VL now supports inference on DCU and XPU.
Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.3.6...v3.3.7
PaddleOCR-VL supports inference using x86-64 CPU.
Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.3.5...v3.3.6
Fixed the issue with weight data type mapping, supporting GPUs with compute capability between 7 and 8.
quantization_config.max_new_tokens parameter during PaddleOCR-VL model inference.quantization_config 时模型解析配置失败的问题。max_new_tokens 参数的传入Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.3.4...v3.3.5
Nothing published for this version
Nothing published for this version
Nothing published for this version
Fix issues such as the missing concatenate_markdown_pages method in the PaddleOCR-VL production pipeline.
concatenate_markdown_pages method in the PaddleOCR-VL production pipeline.concatenate_markdown_pages 方法缺失等问题。Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.3.0...v3.3.1
Added support for inference and deployment of PaddleOCR-VL and PP-OCRv5 multilingual models.
Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.2.1...v3.3.0
…Regular Expression Denial of Service (ReDoS) vulnerability.
Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.2.0...v3.2.1
Deployment Capability Upgrades:
Deployment Capability Upgrades:
Key Model Additions:
Benchmark Enhancements:
Bug Fixes:
Other Updates:
model_name parameter in PaddlePredictorOption has been moved to PaddleInfer, improving usability.部署能力升级:
重要模型新增:
Benchmark升级:
Bug修复:
其他升级:
PaddlePredictorOption 中的 model_name 参数移至 PaddleInfer 中,改善了用户易用性。Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.1.4...v3.2.0
修复了表格识别v2产线在不使用版面区域检测模型时,组batch预测报错的问题。
v3.1.4版本,修复和优化部分问题:
Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.1.3...v3.1.4
修复近期引入的 PP-OCRv4 及以下版本的英文模型推理可视化显示问题。
v3.1.3版本,修复和优化部分问题:
Bug修复:
文档优化:
Full Changelog: https://github.com/PaddlePaddle/PaddleX/compare/v3.1.2...v3.1.3
修复了推理时,当传入不合法后缀名图像文件路径时造成的递归调用报错问题。
v3.1.2版本,修复和优化部分问题:
Bug修复:
功能优化:
文档优化:
v3.1.1版本,修复使用本地字体文件在特殊场景下可能触发的问题
v3.1.1版本,修复使用本地字体文件在特殊场景下可能触发的问题
v3.1.0版本,新增PP-OCRv5种多语种文字识别模型和文档翻译产线,优化PP-StructureV3中的PP-Chart2Table模型:
v3.1.0版本,新增PP-OCRv5种多语种文字识别模型和文档翻译产线,优化PP-StructureV3中的PP-Chart2Table模型:
v3.0.3版本,新增服务化部署多语言调用示例,修复部分问题:
v3.0.3版本,新增服务化部署多语言调用示例,修复部分问题:
功能新增:
Bug修复:
run_mode不生效的问题。page_continuation_flags默认值为None,引发的服务化部署失败问题。v3.0.2版本,新增部分功能,修复和优化部分问题,更新点如下:
v3.0.2版本,新增部分功能,修复和优化部分问题,更新点如下:
功能新增和优化:
BOS改为HuggingFace,同时也支持用户通过更改环境变量PADDLE_PDX_MODEL_SOURCE为BOS,将模型下载源设置为百度云对象存储BOS。Bug修复:
其他:
v3.0.1版本,修复和优化3.0.0版本的部分问题,升级修复点如下:
v3.0.1版本,修复和优化3.0.0版本的部分问题,升级修复点如下:
优化部分模型和模型配置:
limit_side_len由736改为64PP-LCNet_x1_0_textline_ori模型,精度99.42%,OCR、PP-StructureV3、PP-ChatOCRv4产线的默认文本行方向分类器改为该模型PP-LCNet_x0_25_textline_ori模型,精度提升3.3个百分点,当前精度98.85%优化和修复部分问题:
FatalError: Process abort signal is detected by the operating system错误的问题GPUtil不可用时默认使用CPU,同时支持通过CUDA_VISIBLE_DEVICES环境变量控制使用的默认GPU设备PaddleX3.0 汇聚了飞桨多年的精选模型,涵盖多个不同的AI方向,并且在此基础上,统一了模型的接口,降低了模型开发的难度。PaddleX3.0提供了多模型组合使用的方式、提供了较多产业级方便易用的pipeline,并提供了多种模型部署的工具。基于PaddleX3.0,可以快速完成AI应用的开发
PaddleX3.0 汇聚了飞桨多年的精选模型,涵盖多个不同的AI方向,并且在此基础上,统一了模型的接口,降低了模型开发的难度。PaddleX3.0提供了多模型组合使用的方式、提供了较多产业级方便易用的pipeline,并提供了多种模型部署的工具。基于PaddleX3.0,可以快速完成AI应用的开发和落地。PaddleX3.0 一共经历过6个版本,本次发布为PaddleX3.0正式版。相比2.x版本,3.0版本的主要能力如下:
丰富的模型库:
统一推理接口,重构部署能力:
全面适配飞桨框架3.0:
-o Global.dy2st=True 即可开启编译器训练,在 GPU 上,多数模型训练速度可提升 10% 以上,少部分模型训练速度可以提升 30% 以上。推理方面,模型整体适配飞桨 3.0 中间表示技术(PIR),拥有更加灵活的扩展能力和兼容性,静态图模型存储文件名由 xxx.pdmodel 改为 xxx.json。重磅能力支撑:
多硬件支持:
PaddleX 3.0 rc1 全面适配 PaddlePaddle 3.0正式版,核心升级如下:
PaddleX 3.0 rc1 全面适配 PaddlePaddle 3.0正式版,核心升级如下:
全面适配飞桨框架3.0新特性:支持编译器训练,训练命令通过追加 -o Global.dy2st=True 即可开启编译器训练,在 GPU 上,多数模型训练速度可提升 10% 以上,少部分模型训练速度可以提升 30% 以上。推理方面,模型整体适配飞桨 3.0 中间表示技术(PIR),拥有更加灵活的扩展能力和兼容性,静态图模型存储文件名由 xxx.pdmodel 改为 xxx.json。
新增飞桨自研文档图像理解多模态大模型 PP-DocBee:在学术界及内部业务场景文档理解评测榜单上,PP-DocBee 均达到同参数量级别模型的 SOTA 水平。可应用到财报、研报、合同、说明书、法律法规等文档 QA 场景。
全面支持 ONNX 格式模型,支持通过Paddle2ONNX插件转换模型格式。
升级高性能推理:
多硬件支持扩展:
PaddleX 3.0 rc0 全面适配 PaddlePaddle 3.0rc0 及以上版本,新增10+条产线,40+个模型,优化模型和产线API,多硬件适配更多模型。全面升级高性能推理和服务化部署能力。具体新增能力如下:
PaddleX 3.0 rc0 全面适配 PaddlePaddle 3.0rc0 及以上版本,新增10+条产线,40+个模型,优化模型和产线API,多硬件适配更多模型。全面升级高性能推理和服务化部署能力。具体新增能力如下:
新增产线:
新增模型:
模型和产线能力升级:
多硬件支持:
多环境适配
部署能力全面升级:
PaddleX 3.0 Beta2 全面适配 PaddlePaddle 3.0b2 版本。新增通用图像识别、人脸识别、车辆属性识别和行人属性识别产线,同时新增 42 个模型开发全流程适配昇腾 910B,并全面支持[GitHub 站点文档](https://paddlepaddle.github.io
PaddleX 3.0 Beta2 全面适配 PaddlePaddle 3.0b2 版本。新增通用图像识别、人脸识别、车辆属性识别和行人属性识别产线,同时新增 42 个模型开发全流程适配昇腾 910B,并全面支持GitHub 站点文档。 具体新增能力如下:
新增产线:
新增能力:
优化点:
BugFix:
新增超轻量分类模型PPLCNet,在Intel CPU上,单张图像预测速度约5ms,ImageNet-1K数据集上Top1识别准确率达到80.82%,超越ResNet152的模型效果 欢迎体验 新增轻量级检测特色模型PP-PicoDet,第一个在1M参数量之内mAP(0.5:0.95)超越30+(输
新增超轻量分类模型PPLCNet,在Intel CPU上,单张图像预测速度约5ms,ImageNet-1K数据集上Top1识别准确率达到80.82%,超越ResNet152的模型效果 欢迎体验 新增轻量级检测特色模型PP-PicoDet,第一个在1M参数量之内mAP(0.5:0.95)超越30+(输入416像素时),网络预测在ARM CPU下可达150FPS 欢迎体验 升级PaddleX Restful API,支持飞桨动态图开发模式 欢迎体验 新增检测模型负样本训练策略 欢迎体验 新增python轻量级服务化部署 欢迎体验
新增检测任务和实例分割任务的预测结果可视化、以及预测错误原因分析,辅助分析模型效果
Nothing published for this version
Nothing published for this version
目标检测任务新增PP-YOLOv2, COCO test数据集精度达到49.5%、V100预测速度达到68.9 FPS
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
> - 图像分类模型ResNet50_vd新增10万分类预训练模型
模型更新
- 图像分类模型ResNet50_vd新增10万分类预训练模型
- 目标检测模型FasterRCNN新增模型裁剪支持
- 目标检测模型新增多通道图像训练支持
模型部署更新
- 修复OpenVINO部署C++代码中部分Bug
- 树莓派部署新增Arm V8支持
产业案例更新
- 新增工业质检产业案例,提供基于GPU和CPU两种部署场景下的工业质检方案,及与质检相关的优化策略 详情链接
Nothing published for this version
Nothing published for this version
Nothing published for this version
One column per quarter.