[Refine] Fix several bugs on custom device (#1239)

* fix code and docs

* fix more typos

* fix more docs

* fix fourcastnet

* fix ERA5MeteoDataset

* fix more docs

* fix float64 to float32 by default

* remove pass

* fix order

* relax tol for test_montecarlo_integrate

* update paddle_cluster in README

* fix UT

* float64 -> float32

Author: HydrogenSulfate

README.md

@@ -300,6 +300,7 @@ python -c "import paddle; paddle.utils.run_check()"
 | [Open3D](https://github.com/PFCCLab/Open3D.git) | 三维图形库 | 全量支持 |
 | [neuraloperator](https://github.com/PFCCLab/neuraloperator) | 神经算子库 | 全量支持 |
 | [paddle_scatter](https://github.com/PFCCLab/paddle_scatter) | 张量稀疏聚合库 | 全量支持 |
+| [paddle_cluster](https://github.com/PFCCLab/paddle_cluster) | 几何采样聚合库 | 全量支持 |
 | [paddle_sparse](https://github.com/PFCCLab/paddle_sparse) | 张量稀疏计算库 | 全量支持 |
 | [paddle_harmonics](https://github.com/PFCCLab/paddle_harmonics.git) | 球面谐波变换库 | 全量支持 |
 | [deepali](https://github.com/PFCCLab/deepali) | 图像、点云配准库 | 全量支持 |

docs/index.md

@@ -142,7 +142,7 @@
     | 问题类型 | 案例名称 | 优化算法 | 模型类型 | 训练方式 | 数据集 | 参考资料 |
     |-----|---------|-----|---------|----|---------|---------|
     | 材料设计 | [散射板设计(反问题)](./zh/examples/hpinns.md) | 机理驱动 | Transformer | 无监督学习 | [Train Data](https://paddle-org.bj.bcebos.com/paddlescience/datasets/hPINNs/hpinns_holo_train.mat)<br>[Eval Data](https://paddle-org.bj.bcebos.com/paddlescience/datasets/hPINNs/hpinns_holo_valid.mat) | [Paper](https://arxiv.org/pdf/2102.04626.pdf) |
-    | 晶体材料属性预测 | [CGCNN](./zh/examples/cgcnn.md) | 数据驱动 | GNN | 监督学习 | [MP](https://next-gen.materialsproject.org/) / [Perovskite](https://cmr.fysik.dtu.dk/cubic_perovskites/cubic_perovskites.html) / [C2DB](https://cmr.fysik.dtu.dk/c2db/c2db.html) / [test](https://paddle-org.bj.bcebos.com/paddlescience%2Fdatasets%2Fcgcnn%2Fcgcnn-test.zip) | [Paper](https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.120.145301) |
+    | 晶体材料属性预测 | [CGCNN](./zh/examples/cgcnn.md) | 数据驱动 | GNN | 监督学习 | [MP](https://next-gen.materialsproject.org/) / [Perovskite](https://cmr.fysik.dtu.dk/cubic_perovskites/cubic_perovskites.html) / [C2DB](https://cmr.fysik.dtu.dk/c2db/c2db.html) / [test](https://paddle-org.bj.bcebos.com/paddlescience/datasets/cgcnn/cgcnn-test.zip) | [Paper](https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.120.145301) |
     | 二维材料生成与数据库 | [ML2DDB](./en/examples/ml2ddb.md) | 数据驱动 | GNN/Diffusion | 监督学习 | Coming Soon | [Paper](https://arxiv.org/pdf/2507.00584) |
 
 === "地球科学"

docs/zh/api/arch.md

@@ -4,11 +4,12 @@
     handler: python
     options:
       members:
+        - Arch
         - AFNONet
         - AMGNet
-        - Arch
         - AutoEncoder
         - ChipDeepONets
+        - Climateformer
         - CrystalGraphConvNet
         - CuboidTransformer
         - CVit1D
@@ -21,31 +22,30 @@
         - FNO1d
         - Generator
         - HEDeepONets
+        - IFMMLP
         - KAN
         - LatentNO
+        - LNO
         - LorenzEmbedding
+        - Meteoformer
         - MLP
         - ModelList
         - ModifiedMLP
+        - MoleculeModel
         - NowcastNet
+        - Preformer
+        - RegDGCNN
+        - RegPointNet
         - SFNONet
         - SPINN
         - STAFNet
         - TFNO1dNet
         - TFNO2dNet
         - TFNO3dNet
+        - TGCN
         - Transformer
         - UNetEx
         - UNONet
         - USCNN
-        - LNO
-        - TGCN
-        - RegDGCNN
-        - RegPointNet
-        - IFMMLP
-        - Meteoformer
-        - Climateformer
-        - MoleculeModel
-        - Preformer
       show_root_heading: true
       heading_level: 3

docs/zh/api/geometry.md

@@ -11,10 +11,10 @@
         - Hypersphere
         - Interval
         - Mesh
-        - SDFMesh
         - PointCloud
         - Polygon
         - Rectangle
+        - SDFMesh
         - Sphere
         - TimeDomain
         - TimeXGeometry

docs/zh/api/metric.md

@@ -10,10 +10,10 @@
         - LatitudeWeightedACC
         - LatitudeWeightedRMSE
         - MAE
+        - MaxAE
         - MeanL2Rel
         - MSE
-        - RMSE
-        - MaxAE
         - R2Score
+        - RMSE
       show_root_heading: true
       heading_level: 3

docs/zh/examples/CNN_UTS.md

@@ -3,7 +3,7 @@
 === "模型训练命令"
 
     ``` sh
-    python main.py mode=train
+    python main.py
     ```
 
 === "模型评估命令"
@@ -15,10 +15,10 @@
 ## 下载预训练模型
 
 | [resnet18-v5-fold1](https://paddle-org.bj.bcebos.com/paddlescience/models/CNN_UTS/resnet18-v5-fold1.pdparams) |
- [resnet18-v5-fold2](https://paddle-org.bj.bcebos.com/paddlescience/models/CNN_UTS/resnet18-v5-fold1.pdparams) |
- [resnet18-v5-fold3](https://paddle-org.bj.bcebos.com/paddlescience/models/CNN_UTS/resnet18-v5-fold1.pdparams) |
- [resnet18-v5-fold4](https://paddle-org.bj.bcebos.com/paddlescience/models/CNN_UTS/resnet18-v5-fold1.pdparams) |
- [resnet18-v5-fold5](https://paddle-org.bj.bcebos.com/paddlescience/models/CNN_UTS/resnet18-v5-fold1.pdparams) ||
+[resnet18-v5-fold2](https://paddle-org.bj.bcebos.com/paddlescience/models/CNN_UTS/resnet18-v5-fold1.pdparams) |
+[resnet18-v5-fold3](https://paddle-org.bj.bcebos.com/paddlescience/models/CNN_UTS/resnet18-v5-fold1.pdparams) |
+[resnet18-v5-fold4](https://paddle-org.bj.bcebos.com/paddlescience/models/CNN_UTS/resnet18-v5-fold1.pdparams) |
+[resnet18-v5-fold5](https://paddle-org.bj.bcebos.com/paddlescience/models/CNN_UTS/resnet18-v5-fold1.pdparams) |
 
 ## 下载模型必要参数
 
@@ -37,19 +37,19 @@
 ```
 CNN_UTS/
 │
-├─ conf/  
+├─ conf/
 │    └─ resnet.yaml
-├─ data_utils.py  
-├─ model_utils.py  
-├─ main.py  
-├─ requirements.txt  
-├─ readme.md  
-├─ resnet18-v5-finetune/  
-├─ outputs/  
-├─ Saved_Output/  
-└─ Dataset/  
-     ├─ Train_val/  
-     └─ Test/  
+├─ data_utils.py
+├─ model_utils.py
+├─ main.py
+├─ requirements.txt
+├─ readme.md
+├─ resnet18-v5-finetune/
+├─ outputs/
+├─ Saved_Output/
+└─ Dataset/
+     ├─ Train_val/
+     └─ Test/
 ```
 
 ## 2. 模型原理

docs/zh/examples/MLP_LI.md

@@ -12,8 +12,8 @@
 
 | 数据集名称 | 下载链接 |
 |-----------|---------|
-| 训练集 + 验证集 | [MP_data_down_loading(train+validate).csv](https://paddle-org.bj.bcebos.com/paddlescience%2Fdocs%2FMP_data_down_loading(train%2Bvalidate).csv) |
-| 训练集 + 验证集 + 测试集 | [MP_data_down_loading(train+validate+test).csv](https://paddle-org.bj.bcebos.com/paddlescience%2Fdocs%2FMP_data_down_loading(train%2Bvalidate%2Btest).csv) |
+| 训练集 + 验证集 | [MP_data_down_loading(train+validate).csv](https://paddle-org.bj.bcebos.com/paddlescience/docs/MP_data_down_loading(train+validate).csv) |
+| 训练集 + 验证集 + 测试集 | [MP_data_down_loading(train+validate+test).csv](https://paddle-org.bj.bcebos.com/paddlescience/docs/MP_data_down_loading(train+validate+test).csv) |
 
 数据读取需要额外安装依赖 `bayesian-optimization`，请运行安装命令 `pip install bayesian-optimization`。
 
@@ -28,7 +28,7 @@
 
 | 预训练模型                        |
 |-----------------------------------|
-| [MLP_LI_pretrained.pdparams]( https://paddle-org.bj.bcebos.com/paddlescience%2Fmodels%2FMLP_LI_pretrained.pdparams) |
+| [MLP_LI_pretrained.pdparams](https://paddle-org.bj.bcebos.com/paddlescience/models/MLP_LI_pretrained.pdparams) |
 
 
 ## 模型训练命令

docs/zh/examples/cgcnn.md

@@ -11,6 +11,7 @@
     ``` sh
     python CGCNN.py TRAIN_DIR="Your train dataset path" VALID_DIR="Your evaluate dataset path"
     ```
+
 === "模型评估命令"
 
     ``` sh

docs/zh/examples/fourcastnet.md

@@ -41,14 +41,16 @@
 
     ``` sh
     # 下载风速预测小样本数据
-    wget -nc https://paddle-org.bj.bcebos.com/paddlescience%2Fdatasets%2FFourcastNet%2F2018-04-04_n6_precip.npy -P ./datasets/era5/test/
-    wget -nc https://paddle-org.bj.bcebos.com/paddlescience%2Fdatasets%2FFourcastNet%2F2018-04-04_n6.npy -P ./datasets/era5/test/
+    wget -nc https://paddle-org.bj.bcebos.com/paddlescience/datasets/FourcastNet/global_stds.npy -P ./datasets/era5/stat/
+    wget -nc https://paddle-org.bj.bcebos.com/paddlescience/datasets/FourcastNet/global_means.npy -P ./datasets/era5/stat/
+    wget -nc https://paddle-org.bj.bcebos.com/paddlescience/datasets/FourcastNet/2018-04-04_n6_precip.npy -P ./datasets/era5/test/
+    wget -nc https://paddle-org.bj.bcebos.com/paddlescience/datasets/FourcastNet/2018-04-04_n6.npy -P ./datasets/era5/test/
+    # 下载降水量预测小样本数据
+    wget -nc https://paddle-org.bj.bcebos.com/paddlescience/datasets/FourcastNet/2018-09-08_n32.npy -P ./datasets/era5/test/
     # 风速预训练模型推理
     python train_pretrain.py mode=infer
     # 风速微调模型推理
     python train_finetune.py mode=infer
-    # 下载降水量预测小样本数据
-    wget -nc https://paddle-org.bj.bcebos.com/paddlescience%2Fdatasets%2FFourcastNet%2F2018-09-08_n32.npy -P ./datasets/era5/test/
     # 降水量模型推理
     python train_precip.py mode=infer
     ```

docs/zh/examples/ifm.md

@@ -11,7 +11,7 @@
     ``` sh
     # 在tox21/sider/hiv/bace/bbbp等数据上训练模型MLP-IFM,embed_name可选IFM/None
     # mode/data_label/MODEL.embed_name 等参数可在conf/ifm.yaml进行配置
-    python ifm.py mode=train data_label=tox21 MODEL.embed_name='IFM'
+    python ifm.py data_label=tox21 MODEL.embed_name='IFM'
     ```
 
 === "模型评估命令"