【Hackathon 10th Spring No.17】MOFDiff模型复现

[r-cloudforge]

通过迁移 https://github.com/microsoft/MOFDiff 项目实现 MOFDiff 的PaddlePaddle版本复现。

主要改动

1. 新增 ppmat/models/mofdiff/ 模型目录 — 完整的MOFDiff实现，包括VAE编码/解码器和VP扩散过程
2. 新增 structure_generation/configs/mofdiff/ 训练配置
3. 新增 test/mofdiff/ 测试代码 — 18个单元测试

模型特性

• VAE + VP (Variance Preserving) 扩散模型，用于金属有机框架(MOF)结构生成
• SimpleGNN编码器/解码器用于图结构处理
• 高斯傅里叶投影时间嵌入
• VP噪声调度器，支持前向SDE和反向ODE
• 拓扑感知的连续和离散特征处理
• 基于PaddleMaterials标准API：forward(data) → {"loss_dict", "pred_dict"}
• sample() 方法支持完整采样流程
• 已注册到 ppmat/models/__init__.py

测试

cd test && python -m pytest mofdiff/ -v
# 18 passed

参考链接

• 原始论文: https://arxiv.org/abs/2310.10732
• 原始仓库: https://github.com/microsoft/MOFDiff (MIT License)
• Hackathon Issue: 【Hackathon 10th】开源贡献个人挑战赛 · 春节特别季 Paddle#77429

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

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.}

[paddle-bot]

Thanks for your contribution!