Add export & inference for hPINNs

docs/zh/examples/hpinns.md

@@ -26,6 +26,26 @@
     python holography.py mode=eval EVAL.pretrained_model_path=https://paddle-org.bj.bcebos.com/paddlescience/models/hPINNs/hpinns_pretrained.pdparams
     ```
 
+=== "模型导出命令"
+
+    ``` sh
+    python holography.py mode=export
+    ```
+
+=== "模型推理命令"
+
+    ``` sh
+    # linux
+    wget -nc https://paddle-org.bj.bcebos.com/paddlescience/datasets/hPINNs/hpinns_holo_train.mat -P ./datasets/
+    wget -nc https://paddle-org.bj.bcebos.com/paddlescience/datasets/hPINNs/hpinns_holo_valid.mat -P ./datasets/
+    # windows
+    # curl https://paddle-org.bj.bcebos.com/paddlescience/datasets/hPINNs/hpinns_holo_train.mat --output ./datasets/hpinns_holo_train.mat
+    # curl https://paddle-org.bj.bcebos.com/paddlescience/datasets/hPINNs/hpinns_holo_valid.mat --output ./datasets/hpinns_holo_valid.mat
+    python holography.py mode=infer
+    ```
+
+
+
 | 预训练模型  | 指标 |
 |:--| :--|
 | [hpinns_pretrained.pdparams](https://paddle-org.bj.bcebos.com/paddlescience/models/hPINNs/hpinns_pretrained.pdparams) | loss(opt_sup): 0.05352<br>MSE.eval_metric(opt_sup): 0.00002<br>loss(val_sup): 0.02205<br>MSE.eval_metric(val_sup): 0.00001 |

examples/hpinns/conf/hpinns.yaml

@@ -25,6 +25,7 @@ seed: 42
 output_dir: ${hydra:run.dir}
 DATASET_PATH: ./datasets/hpinns_holo_train.mat
 DATASET_PATH_VALID: ./datasets/hpinns_holo_valid.mat
+log_freq: 20
 
 # set working condition
 TRAIN_MODE: aug_lag  # "soft", "penalty", "aug_lag"
@@ -65,3 +66,22 @@ TRAIN:
 # evaluation settings
 EVAL:
   pretrained_model_path: null
+
+# inference settings
+INFER:
+  pretrained_model_path: "https://paddle-org.bj.bcebos.com/paddlescience/models/hPINNs/hpinns_pretrained.pdparams"
+  export_path: ./inference/hpinns
+  pdmodel_path: ${INFER.export_path}.pdmodel
+  pdiparams_path: ${INFER.export_path}.pdiparams
+  output_keys: ["e_re", "e_im", "eps"]
+  device: gpu
+  engine: native
+  precision: fp32
+  onnx_path: ${INFER.export_path}.onnx
+  ir_optim: true
+  min_subgraph_size: 10
+  gpu_mem: 8000
+  gpu_id: 0
+  batch_size: 128
+  max_batch_size: 128
+  num_cpu_threads: 4

examples/hpinns/holography.py

@@ -409,14 +409,79 @@ def evaluate(cfg: DictConfig):
     solver.eval()
 
 
+def export(cfg: DictConfig):
+    # set model
+    model_re = ppsci.arch.MLP(**cfg.MODEL.re_net)
+    model_im = ppsci.arch.MLP(**cfg.MODEL.im_net)
+    model_eps = ppsci.arch.MLP(**cfg.MODEL.eps_net)
+
+    # register transform
+    model_re.register_input_transform(func_module.transform_in)
+    model_im.register_input_transform(func_module.transform_in)
+    model_eps.register_input_transform(func_module.transform_in)
+
+    model_re.register_output_transform(func_module.transform_out_real_part)
+    model_im.register_output_transform(func_module.transform_out_imaginary_part)
+    model_eps.register_output_transform(func_module.transform_out_epsilon)
+
+    # wrap to a model_list
+    model_list = ppsci.arch.ModelList((model_re, model_im, model_eps))
+
+    # initialize solver
+    solver = ppsci.solver.Solver(
+        model_list,
+        pretrained_model_path=cfg.INFER.pretrained_model_path,
+    )
+
+    # export model
+    from paddle.static import InputSpec
+
+    input_spec = [
+        {key: InputSpec([None, 1], "float32", name=key) for key in ["x", "y"]},
+    ]
+    solver.export(input_spec, cfg.INFER.export_path)
+
+
+def inference(cfg: DictConfig):
+    from deploy.python_infer import pinn_predictor
+
+    predictor = pinn_predictor.PINNPredictor(cfg)
+
+    valid_dict = ppsci.utils.reader.load_mat_file(
+        cfg.DATASET_PATH_VALID, ("x_val", "y_val", "bound")
+    )
+    input_dict = {"x": valid_dict["x_val"], "y": valid_dict["y_val"]}
+
+    output_dict = predictor.predict(input_dict, cfg.INFER.batch_size)
+
+    # mapping data to cfg.INFER.output_keys
+    output_dict = {
+        store_key: output_dict[infer_key]
+        for store_key, infer_key in zip(cfg.INFER.output_keys, output_dict.keys())
+    }
+
+    ppsci.visualize.save_vtu_from_dict(
+        "./hpinns_pred.vtu",
+        {**input_dict, **output_dict},
+        input_dict.keys(),
+        cfg.INFER.output_keys,
+    )
+
+
 @hydra.main(version_base=None, config_path="./conf", config_name="hpinns.yaml")
 def main(cfg: DictConfig):
     if cfg.mode == "train":
         train(cfg)
     elif cfg.mode == "eval":
         evaluate(cfg)
+    elif cfg.mode == "export":
+        export(cfg)
+    elif cfg.mode == "infer":
+        inference(cfg)
     else:
-        raise ValueError(f"cfg.mode should in ['train', 'eval'], but got '{cfg.mode}'")
+        raise ValueError(
+            f"cfg.mode should in ['train', 'eval', 'export', 'infer'], but got '{cfg.mode}'"
+        )
 
 
 if __name__ == "__main__":