Skip to content

【Hackathon 10th Spring No.16】RFC: Add DiffSyn model to PaddleMaterials#1305

Open
danqing-li wants to merge 3 commits intoPaddlePaddle:masterfrom
CloudForge-Solutions:task/hackathon10-rfc-016-diffsyn
Open

【Hackathon 10th Spring No.16】RFC: Add DiffSyn model to PaddleMaterials#1305
danqing-li wants to merge 3 commits intoPaddlePaddle:masterfrom
CloudForge-Solutions:task/hackathon10-rfc-016-diffsyn

Conversation

@danqing-li
Copy link
Copy Markdown

@danqing-li danqing-li commented Apr 7, 2026

概述

本 RFC 提出将 DiffSyn(基于扩散模型的沸石合成条件生成方法)模型复现并集成到 PaddleMaterials 中。

主要内容

  1. 将 DiffSyn 条件扩散模型从 PyTorch 迁移至 Paddle,集成到 ppmat/models/diffsyn/,包含沸石 3D 拓扑图 GNN 编码器、OSDA 2D 分子图 GNN 编码器、合成参数空间条件扩散去噪网络
  2. 支持 ZeoSyn 数据集加载(23,961 条沸石合成路线,233 种拓扑,921 种 OSDA)
  3. 实现 classifier-free guidance 采样,训练时以概率随机 drop 条件向量,推理时用引导系数控制生成质量
  4. 适配 PaddleMaterials 统一的 trainer/predictor/sampler 模式
  5. 复现论文中的生成质量指标(Wasserstein 距离 ~0.423,MAE 误差 ±5% 以内)
  6. 新建 synthesis_condition_prediction/ 任务目录,填补 PaddleMaterials 在逆向合成设计领域的空白

技术特点

  • DiffSyn 在连续合成参数空间(结晶温度、H₂O/T 比、结晶时间、gel 组成)上进行条件扩散去噪,而非原子坐标扩散
  • 双 GNN 条件编码器架构:SchNet3D 编码沸石 3D 拓扑结构 + GIN 编码 OSDA 2D 分子图,拼接后作为扩散模型条件输入
  • 使用标准 PyTorch + PyG 操作,无自定义 CUDA kernel,torch→Paddle 迁移所有核心 API 均有对应实现
  • PaddleMaterials 首个合成条件预测模型,建立从目标结构反推合成配方的逆向合成范式

排期

3~4周完成(环境搭建 → GNN 编码器迁移 → 条件扩散模型 → 训练对齐 → WSD/MAE 评估验证 → 文档 PR)

@paddle-bot
Copy link
Copy Markdown

paddle-bot bot commented Apr 7, 2026

你的PR提交成功,感谢你对开源项目的贡献!
请检查PR提交格式和内容是否完备,具体请参考示例模版
Your PR has been submitted. Thanks for your contribution!
Please check its format and content. For this, you can refer to Template and Demo.

@danqing-li danqing-li changed the title 【Hackathon 10th Spring No.】RFC: Add DiffSyn model to PaddleMaterials (Tas… 【Hackathon 10th Spring No.16】RFC: Add DiffSyn model to PaddleMaterials Apr 7, 2026
@CLAassistant
Copy link
Copy Markdown

CLA assistant check
Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
1 out of 2 committers have signed the CLA.

✅ danqing-li
❌ cloudforge1


cloudforge1 seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
You have signed the CLA already but the status is still pending? Let us recheck it.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants