Upload folder using huggingface_hub

README.md

@@ -0,0 +1,118 @@
+<p align="center">
+  <strong>
+    <span style="font-size: 30px;">NEP</span>
+  </strong>
+</p>
+
+# Model Introduction
+
+NEP (Neural Evolution Potential) is a neural-network potential training example based on MatPL. It learns material interaction information such as energy, forces, and virial from atomic structure data.
+
+
+# Model Description
+
+NEP is based on the MatPL framework and is trained on atomic-system data such as Cu and LiSiC. It performs interatomic potential function training and molecular dynamics simulation for materials systems.
+
+# Applicable Scenarios
+
+| Scenario | Description |
+| :---: | :--- |
+| Cu system NEP training | Train the Cu potential using `demo/nep_Cu/Cu_nep_train.json` |
+| LiSiC system NEP training | Train the LiSiC potential using `demo/nep_LiSiC/LiSiC_nep_train.json` |
+| SLURM job submission | Refer to `demo/nep_Cu/submit.sh` to run training on a cluster |
+| Custom data migration | Organize your data into MatPL-supported formats such as `pwmat/movement` and replace the training path |
+
+# Usage Instructions
+
+## 1. Using OneCode
+
+You can try out intelligent one-click AI4S programming in the OneCode online environment:
+
+[Try intelligent one-click AI4S programming](https://web-2069360198568017922-iaaj.ksai.scnet.cn:58043/home)
+
+## 2. Manual Installation and Usage
+
+**Hardware Requirements**
+
+- DCU or GPU is recommended for training.
+- CPU can be used for import and small-configuration connectivity checks; full training will be slow.
+- DCU users need to install DTK in advance. DTK 25.04.2 or above, or the OneScience-recommended version matching the current cluster, is suggested.
+
+### Download the Model Package
+
+```bash
+modelscope download --model OneScience/NEP --local_dir ./nep
+cd nep
+```
+
+### Install the Runtime Environment
+
+**DCU Environment**
+
+```bash
+# Please activate DTK and CONDA first
+conda create -n onescience311 python=3.11 -y
+conda activate onescience311
+# uv installation is also supported
+pip install onescience[matchem-dcu] -i http://mirrors.onescience.ai:3141/pypi/simple/  --trusted-host mirrors.onescience.ai
+```
+
+**GPU Environment**
+
+```bash
+# Please activate CONDA first
+conda create -n onescience311 python=3.11 -y libstdcxx-ng=12 libgcc-ng=12 gcc_linux-64=12 gxx_linux-64=12
+conda activate onescience311
+# uv installation is also supported
+pip install onescience[matchem-gpu] -i http://mirrors.onescience.ai:3141/pypi/simple/  --trusted-host mirrors.onescience.ai
+```
+
+### Install MatPL
+
+```bash
+# Uses the test_pip environment by default; if you use another conda environment name, please specify it first:
+# export MATCHEM_CONDA_NAME=your_env
+bash matpl_install.sh
+```
+
+### Training Data Description
+
+This repository does not include built-in training data. Taking the MatPL dataset as an example, download it from ModelScope and place it under `data/` in the repository root:
+
+```bash
+modelscope download --dataset OneScience/MatPL --local_dir ./data
+```
+
+After downloading, the data path will be `data/MatPL/`.
+
+### Training
+
+Single-GPU:
+
+```bash
+cd demo/nep_Cu
+MatPL train Cu_nep_train.json
+```
+
+SLURM submission:
+
+```bash
+cd demo/nep_Cu
+bash submit.sh
+```
+
+### Training Weights
+
+This repository currently does not include built-in trained weights. Weights can be obtained through the training steps above.
+
+# OneScience Official Information
+
+| Platform | OneScience Main Repository | Skills Repository |
+| --- | --- | --- |
+| Gitee | https://gitee.com/onescience-ai/onescience | https://gitee.com/onescience-ai/oneskills |
+| GitHub | https://github.com/onescience-ai/OneScience | https://github.com/onescience-ai/oneskills |
+
+# Citation and License
+
+- The NEP example code comes from the OneScience repository. This repository retains the source attribution and is organized for OneScience ModelScope automatic execution scenarios.
+- If you use NEP or MatPL training results in scientific research, we recommend citing the relevant OneScience project information, MatPL/NEP methods, and the sources of the datasets actually used.

demo/nep_AuAg/dcu_1/AuAg_nep_train.json

@@ -0,0 +1,80 @@
+{
+    "model_type": "NEP",
+    "atom_type": [
+        "Ag",
+        "Au"
+    ],
+    "seed": 2023,
+    "recover_train": true,
+    "model": {
+        "descriptor": {
+            "cutoff": [
+                6.0,
+                4.0
+            ],
+            "n_max": [
+                16,
+                12
+            ],
+            "basis_size": [
+                12,
+                8
+            ],
+            "l_max": [
+                4,
+                2,
+                1
+            ]
+        },
+        "fitting_net": {
+            "network_size": [
+                60,
+                1
+            ]
+        }
+    },
+    "optimizer": {
+        "optimizer": "ADAM",
+        "epochs": 30,
+        "batch_size": 8,
+        "print_freq": 10,
+        "learning_rate": 0.001,
+        "stop_lr": 3.51e-08,
+        "stop_step": 1000000,
+        "decay_step": 5000,
+        "train_energy": true,
+        "train_force": true,
+        "train_virial": true,
+        "start_pre_fac_force": 1000,
+        "start_pre_fac_etot": 0.02,
+        "start_pre_fac_virial": 50.0,
+        "end_pre_fac_force": 1.0,
+        "end_pre_fac_etot": 1.0,
+        "end_pre_fac_virial": 1.0,
+        "block_size": 5120,
+        "kalman_lambda": 0.98,
+        "kalman_nue": 0.9987,
+        "pre_fac_force": 2.0,
+        "pre_fac_etot": 1.0,
+        "pre_fac_virial": 1.0,
+        "max_norm": 0.5,
+        "norm_type": 2,
+        "t_0": 1,
+        "t_mult": 2,
+        "verbose": true,
+        "_clip_value": 1.0
+    },
+    "format": "pwmlff/npy",
+    "train_data": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/pwdata/Ag10Au22",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/pwdata/Ag10Au38",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/pwdata/Ag10Au6",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/pwdata/Ag11Au21",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/pwdata/Ag11Au37",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/pwdata/Ag11Au5",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/pwdata/Ag12"
+    ],
+    "valid_data": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/pwdata/Ag12"
+    ]
+}

demo/nep_AuAg/dcu_1/std_input.json

@@ -0,0 +1,72 @@
+{
+    "model_type": "NEP",
+    "atom_type": [
+        47,
+        79
+    ],
+    "seed": 2023,
+    "recover_train": true,
+    "model": {
+        "descriptor": {
+            "cutoff": [
+                6.0,
+                4.0
+            ],
+            "n_max": [
+                16,
+                12
+            ],
+            "basis_size": [
+                12,
+                8
+            ],
+            "l_max": [
+                4,
+                2,
+                1
+            ]
+        },
+        "fitting_net": {
+            "network_size": [
+                60,
+                1
+            ]
+        }
+    },
+    "optimizer": {
+        "optimizer": "ADAM",
+        "epochs": 30,
+        "batch_size": 8,
+        "print_freq": 10,
+        "learning_rate": 0.001,
+        "stop_lr": 3.51e-08,
+        "stop_step": 1000000,
+        "decay_step": 5000,
+        "train_energy": true,
+        "train_force": true,
+        "train_virial": true,
+        "start_pre_fac_force": 1000,
+        "start_pre_fac_etot": 0.02,
+        "start_pre_fac_virial": 50.0,
+        "end_pre_fac_force": 1.0,
+        "end_pre_fac_etot": 1.0,
+        "end_pre_fac_virial": 1.0,
+        "max_norm": 0.5,
+        "norm_type": 2,
+        "t_0": 1,
+        "t_mult": 2
+    },
+    "format": "pwmlff/npy",
+    "train_data": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/pwdata/Ag10Au22",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/pwdata/Ag10Au38",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/pwdata/Ag10Au6",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/pwdata/Ag11Au21",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/pwdata/Ag11Au37",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/pwdata/Ag11Au5",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/pwdata/Ag12"
+    ],
+    "valid_data": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/pwdata/Ag12"
+    ]
+}

demo/nep_AuAg/dcu_1/submit.sh

@@ -0,0 +1,50 @@
+#!/bin/bash
+#SBATCH --job-name=nep_AuAg_train
+#SBATCH --partition=hx1hdexclu12
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=1
+#SBATCH --gres=dcu:1
+#SBATCH --cpus-per-task=16
+#SBATCH --time=2:00:00
+#SBATCH --output=slurm_%j.out
+#SBATCH --error=slurm_%j.err
+
+if [[ -n "${SCRIPT_DIR}" ]]; then
+    :
+elif [[ -n "${SLURM_SUBMIT_DIR}" ]]; then
+    SCRIPT_DIR="${SLURM_SUBMIT_DIR}"
+else
+    echo "ERROR: 未检测到外部 SCRIPT_DIR 变量，也不在 Slurm 任务环境（SLURM_SUBMIT_DIR 为空）"
+    exit 1
+fi
+echo $SCRIPT_DIR
+
+export MATCHEM_CONDA_NAME="${MATCHEM_CONDA_NAME:-test_pip}"
+source "$SCRIPT_DIR/matchem_env.sh"
+
+# MatPL 运行时环境（不侵入 matchem_env.sh，各算例自行维护）
+# 校验MatPL根路径
+if [[ -z "${MATPL_SRC_DIR}" || "${MATPL_SRC_DIR}" == "/path/to/matpl_dcu" || ! -d "${MATPL_SRC_DIR}" ]]; then
+    echo "=============================================="
+    echo "ERROR: MATPL_SRC_DIR 路径配置错误！"
+    echo "当前值：${MATPL_SRC_DIR}"
+    echo "请设置为真实有效的MatPL源码目录，示例："
+    echo "export MATPL_SRC_DIR=/public/home/xxx/real_matpl_path"
+    echo "=============================================="
+    exit 1
+fi
+
+if [ -f "$MATPL_SRC_DIR/env.sh" ]; then
+    source "$MATPL_SRC_DIR/env.sh"
+fi
+echo "MATPL_SRC_DIR = $MATPL_SRC_DIR"
+export LD_LIBRARY_PATH="$MATPL_SRC_DIR/src/op/build/lib:${LD_LIBRARY_PATH:-}"
+
+
+# 输入 JSON 使用环境变量，运行前展开为临时文件
+INPUT_JSON="AuAg_nep_train.json"
+EXPANDED_JSON=".${INPUT_JSON%.json}_expanded_$$.json"
+trap 'rm -f "$EXPANDED_JSON"' EXIT
+python3 -c "import os; open('$EXPANDED_JSON','w').write(os.path.expandvars(open('$INPUT_JSON').read()))"
+
+MatPL train "$EXPANDED_JSON"

demo/nep_AuAg/dcu_16/AuAg_nep_train_16card.json

@@ -0,0 +1,51 @@
+{
+    "model_type": "NEP",
+    "atom_type": ["Ag", "Au"],
+    "seed": 2023,
+    "recover_train": true,
+    "model": {
+        "descriptor": {
+            "cutoff": [6.0, 4.0],
+            "n_max": [16, 12],
+            "basis_size": [12, 8],
+            "l_max": [4, 2, 1]
+        },
+        "fitting_net": {
+            "network_size": [100, 1]
+        }
+    },
+    "optimizer": {
+        "optimizer": "ADAM",
+        "epochs": 30,
+        "batch_size": 128,
+        "print_freq": 10,
+        "learning_rate": 0.001,
+        "stop_lr": 3.51e-08,
+        "stop_step": 1000000,
+        "decay_step": 5000,
+        "train_energy": true,
+        "train_force": true,
+        "train_virial": true,
+        "start_pre_fac_force": 1000,
+        "start_pre_fac_etot": 0.02,
+        "start_pre_fac_virial": 50.0,
+        "end_pre_fac_force": 1.0,
+        "end_pre_fac_etot": 1.0,
+        "end_pre_fac_virial": 1.0,
+        "block_size": 5120,
+        "kalman_lambda": 0.98,
+        "kalman_nue": 0.9987,
+        "pre_fac_force": 2.0,
+        "pre_fac_etot": 1.0,
+        "pre_fac_virial": 1.0,
+        "max_norm": 0.5,
+        "norm_type": 2,
+        "t_0": 1,
+        "t_mult": 2,
+        "verbose": true,
+        "_clip_value": 1.0
+    },
+    "format": "extxyz",
+    "train_data": ["/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/AuAg-5762.xyz"],
+    "valid_data": ["/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/AuAg-5762.xyz"]
+}

demo/nep_AuAg/dcu_16/submit_16card.sh

@@ -0,0 +1,61 @@
+#!/bin/bash
+#SBATCH --job-name=nep_AuAg_16card
+#SBATCH --partition=hx1hdexclu12
+#SBATCH --nodes=2
+#SBATCH --ntasks-per-node=8
+#SBATCH --gres=dcu:8
+#SBATCH --cpus-per-task=4
+#SBATCH --mem=256G
+#SBATCH --time=2:00:00
+#SBATCH --output=slurm_16card_%j.out
+#SBATCH --error=slurm_16card_%j.err
+
+if [[ -n "${SCRIPT_DIR}" ]]; then
+    :
+elif [[ -n "${SLURM_SUBMIT_DIR}" ]]; then
+    SCRIPT_DIR="${SLURM_SUBMIT_DIR}"
+else
+    echo "ERROR: 未检测到外部 SCRIPT_DIR 变量，也不在 Slurm 任务环境（SLURM_SUBMIT_DIR 为空）"
+    exit 1
+fi
+echo $SCRIPT_DIR
+
+export MATCHEM_CONDA_NAME="${MATCHEM_CONDA_NAME:-test_pip}"
+source "$SCRIPT_DIR/matchem_env.sh"
+
+# MatPL 运行时环境（不侵入 matchem_env.sh，各算例自行维护）
+# 校验MatPL根路径
+if [[ -z "${MATPL_SRC_DIR}" || "${MATPL_SRC_DIR}" == "/path/to/matpl_dcu" || ! -d "${MATPL_SRC_DIR}" ]]; then
+    echo "=============================================="
+    echo "ERROR: MATPL_SRC_DIR 路径配置错误！"
+    echo "当前值：${MATPL_SRC_DIR}"
+    echo "请设置为真实有效的MatPL源码目录，示例："
+    echo "export MATPL_SRC_DIR=/public/home/xxx/real_matpl_path"
+    echo "=============================================="
+    exit 1
+fi
+
+if [ -f "$MATPL_SRC_DIR/env.sh" ]; then
+    source "$MATPL_SRC_DIR/env.sh"
+fi
+echo "MATPL_SRC_DIR = $MATPL_SRC_DIR"
+export LD_LIBRARY_PATH="$MATPL_SRC_DIR/src/op/build/lib:${LD_LIBRARY_PATH:-}"
+
+# 多节点通信设置
+MASTER_ADDR=$(scontrol show hostnames $SLURM_JOB_NODELIST | head -n 1)
+function get_free_port() {
+    python -c 'import socket; s = socket.socket(socket.AF_INET, socket.SOCK_STREAM); s.bind(("", 0)); print(s.getsockname()[1]); s.close()'
+}
+MASTER_PORT=$(get_free_port)
+export MASTER_ADDR=$MASTER_ADDR
+export MASTER_PORT=$MASTER_PORT
+
+# 16 卡训练（2 节点 × 8 DCU）
+
+# 输入 JSON 使用环境变量，运行前展开为临时文件
+INPUT_JSON="AuAg_nep_train_16card.json"
+EXPANDED_JSON=".${INPUT_JSON%.json}_expanded_$$.json"
+trap 'rm -f "$EXPANDED_JSON"' EXIT
+python3 -c "import os; open('$EXPANDED_JSON','w').write(os.path.expandvars(open('$INPUT_JSON').read()))"
+
+srun MatPL train "$EXPANDED_JSON"

demo/nep_AuAg/dcu_8/AuAg_nep_train_8card.json

@@ -0,0 +1,51 @@
+{
+    "model_type": "NEP",
+    "atom_type": ["Ag", "Au"],
+    "seed": 2023,
+    "recover_train": true,
+    "model": {
+        "descriptor": {
+            "cutoff": [6.0, 4.0],
+            "n_max": [16, 12],
+            "basis_size": [12, 8],
+            "l_max": [4, 2, 1]
+        },
+        "fitting_net": {
+            "network_size": [100, 1]
+        }
+    },
+    "optimizer": {
+        "optimizer": "ADAM",
+        "epochs": 30,
+        "batch_size": 64,
+        "print_freq": 10,
+        "learning_rate": 0.001,
+        "stop_lr": 3.51e-08,
+        "stop_step": 1000000,
+        "decay_step": 5000,
+        "train_energy": true,
+        "train_force": true,
+        "train_virial": true,
+        "start_pre_fac_force": 1000,
+        "start_pre_fac_etot": 0.02,
+        "start_pre_fac_virial": 50.0,
+        "end_pre_fac_force": 1.0,
+        "end_pre_fac_etot": 1.0,
+        "end_pre_fac_virial": 1.0,
+        "block_size": 5120,
+        "kalman_lambda": 0.98,
+        "kalman_nue": 0.9987,
+        "pre_fac_force": 2.0,
+        "pre_fac_etot": 1.0,
+        "pre_fac_virial": 1.0,
+        "max_norm": 0.5,
+        "norm_type": 2,
+        "t_0": 1,
+        "t_mult": 2,
+        "verbose": true,
+        "_clip_value": 1.0
+    },
+    "format": "extxyz",
+    "train_data": ["/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/AuAg-5762.xyz"],
+    "valid_data": ["/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/AuAg/AuAg-5762.xyz"]
+}

demo/nep_AuAg/dcu_8/submit_8card.sh

@@ -0,0 +1,51 @@
+#!/bin/bash
+#SBATCH --job-name=nep_AuAg_8card
+#SBATCH --partition=hx1hdexclu12
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=8
+#SBATCH --gres=dcu:8
+#SBATCH --cpus-per-task=4
+#SBATCH --mem=256G
+#SBATCH --time=2:00:00
+#SBATCH --output=slurm_8card_%j.out
+#SBATCH --error=slurm_8card_%j.err
+
+if [[ -n "${SCRIPT_DIR}" ]]; then
+    :
+elif [[ -n "${SLURM_SUBMIT_DIR}" ]]; then
+    SCRIPT_DIR="${SLURM_SUBMIT_DIR}"
+else
+    echo "ERROR: 未检测到外部 SCRIPT_DIR 变量，也不在 Slurm 任务环境（SLURM_SUBMIT_DIR 为空）"
+    exit 1
+fi
+echo $SCRIPT_DIR
+
+export MATCHEM_CONDA_NAME="${MATCHEM_CONDA_NAME:-test_pip}"
+source "$SCRIPT_DIR/matchem_env.sh"
+
+# MatPL 运行时环境（不侵入 matchem_env.sh，各算例自行维护）
+# 校验MatPL根路径
+if [[ -z "${MATPL_SRC_DIR}" || "${MATPL_SRC_DIR}" == "/path/to/matpl_dcu" || ! -d "${MATPL_SRC_DIR}" ]]; then
+    echo "=============================================="
+    echo "ERROR: MATPL_SRC_DIR 路径配置错误！"
+    echo "当前值：${MATPL_SRC_DIR}"
+    echo "请设置为真实有效的MatPL源码目录，示例："
+    echo "export MATPL_SRC_DIR=/public/home/xxx/real_matpl_path"
+    echo "=============================================="
+    exit 1
+fi
+
+if [ -f "$MATPL_SRC_DIR/env.sh" ]; then
+    source "$MATPL_SRC_DIR/env.sh"
+fi
+echo "MATPL_SRC_DIR = $MATPL_SRC_DIR"
+export LD_LIBRARY_PATH="$MATPL_SRC_DIR/src/op/build/lib:${LD_LIBRARY_PATH:-}"
+
+# 8 卡训练
+# 输入 JSON 使用环境变量，运行前展开为临时文件
+INPUT_JSON="AuAg_nep_train_8card.json"
+EXPANDED_JSON=".${INPUT_JSON%.json}_expanded_$$.json"
+trap 'rm -f "$EXPANDED_JSON"' EXIT
+python3 -c "import os; open('$EXPANDED_JSON','w').write(os.path.expandvars(open('$INPUT_JSON').read()))"
+
+MatPL train "$EXPANDED_JSON"

demo/nep_Cu/Cu_nep_train.json

@@ -0,0 +1,62 @@
+{
+    "model_type": "NEP",
+    "atom_type": [
+        29
+    ],
+    "seed": 2023,
+    "recover_train": true,
+    "model": {
+        "descriptor": {
+            "cutoff": [
+                6.0,
+                6.0
+            ],
+            "n_max": [
+                4,
+                4
+            ],
+            "basis_size": [
+                12,
+                12
+            ],
+            "l_max": [
+                4,
+                2,
+                1
+            ]
+        },
+        "fitting_net": {
+            "network_size": [
+                40,
+                1
+            ]
+        }
+    },
+    "optimizer": {
+        "optimizer": "ADAM",
+        "epochs": 30,
+        "batch_size": 1,
+        "print_freq": 10,
+        "learning_rate": 0.001,
+        "stop_lr": 3.51e-08,
+        "stop_step": 1000000,
+        "decay_step": 5000,
+        "train_energy": true,
+        "train_force": true,
+        "train_virial": false,
+        "start_pre_fac_force": 1000,
+        "start_pre_fac_etot": 0.02,
+        "start_pre_fac_virial": 50.0,
+        "end_pre_fac_force": 1.0,
+        "end_pre_fac_etot": 1.0,
+        "end_pre_fac_virial": 1.0
+    },
+    "format": "pwmat/movement",
+    "train_data": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/Cu/pwdata/0_300_MOVEMENT",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/Cu/pwdata/1_500_MOVEMENT"
+    ],
+    "valid_data": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/Cu/pwdata/valid_movement"
+    ]
+}

demo/nep_Cu/std_input.json

@@ -0,0 +1,62 @@
+{
+    "model_type": "NEP",
+    "atom_type": [
+        29
+    ],
+    "seed": 2023,
+    "recover_train": true,
+    "model": {
+        "descriptor": {
+            "cutoff": [
+                6.0,
+                6.0
+            ],
+            "n_max": [
+                4,
+                4
+            ],
+            "basis_size": [
+                12,
+                12
+            ],
+            "l_max": [
+                4,
+                2,
+                1
+            ]
+        },
+        "fitting_net": {
+            "network_size": [
+                40,
+                1
+            ]
+        }
+    },
+    "optimizer": {
+        "optimizer": "ADAM",
+        "epochs": 30,
+        "batch_size": 1,
+        "print_freq": 10,
+        "learning_rate": 0.001,
+        "stop_lr": 3.51e-08,
+        "stop_step": 1000000,
+        "decay_step": 5000,
+        "train_energy": true,
+        "train_force": true,
+        "train_virial": false,
+        "start_pre_fac_force": 1000,
+        "start_pre_fac_etot": 0.02,
+        "start_pre_fac_virial": 50.0,
+        "end_pre_fac_force": 1.0,
+        "end_pre_fac_etot": 1.0,
+        "end_pre_fac_virial": 1.0
+    },
+    "format": "pwmat/movement",
+    "train_data": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/Cu/pwdata/0_300_MOVEMENT",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/Cu/pwdata/1_500_MOVEMENT"
+    ],
+    "valid_data": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/Cu/pwdata/valid_movement"
+    ]
+}

demo/nep_Cu/submit.sh

@@ -0,0 +1,49 @@
+#!/bin/bash
+#SBATCH --job-name=nep_Cu_train
+#SBATCH --partition=hx1hdexclu12
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=1
+#SBATCH --gres=dcu:1
+#SBATCH --cpus-per-task=16
+#SBATCH --time=2:00:00
+#SBATCH --output=slurm_%j.out
+#SBATCH --error=slurm_%j.err
+
+if [[ -n "${SCRIPT_DIR}" ]]; then
+    :
+elif [[ -n "${SLURM_SUBMIT_DIR}" ]]; then
+    SCRIPT_DIR="${SLURM_SUBMIT_DIR}"
+else
+    echo "ERROR: 未检测到外部 SCRIPT_DIR 变量，也不在 Slurm 任务环境（SLURM_SUBMIT_DIR 为空）"
+    exit 1
+fi
+echo $SCRIPT_DIR
+
+export MATCHEM_CONDA_NAME="${MATCHEM_CONDA_NAME:-test_pip}"
+source "$SCRIPT_DIR/matchem_env.sh"
+
+# MatPL 运行时环境（不侵入 matchem_env.sh，各算例自行维护）
+# 校验MatPL根路径
+if [[ -z "${MATPL_SRC_DIR}" || "${MATPL_SRC_DIR}" == "/path/to/matpl_dcu" || ! -d "${MATPL_SRC_DIR}" ]]; then
+    echo "=============================================="
+    echo "ERROR: MATPL_SRC_DIR 路径配置错误！"
+    echo "当前值：${MATPL_SRC_DIR}"
+    echo "请设置为真实有效的MatPL源码目录，示例："
+    echo "export MATPL_SRC_DIR=/public/home/xxx/real_matpl_path"
+    echo "=============================================="
+    exit 1
+fi
+
+if [ -f "$MATPL_SRC_DIR/env.sh" ]; then
+    source "$MATPL_SRC_DIR/env.sh"
+fi
+echo "MATPL_SRC_DIR = $MATPL_SRC_DIR"
+export LD_LIBRARY_PATH="$MATPL_SRC_DIR/src/op/build/lib:${LD_LIBRARY_PATH:-}"
+
+# 输入 JSON 使用环境变量，运行前展开为临时文件
+INPUT_JSON="Cu_nep_train.json"
+EXPANDED_JSON=".${INPUT_JSON%.json}_expanded_$$.json"
+trap 'rm -f "$EXPANDED_JSON"' EXIT
+python3 -c "import os; open('$EXPANDED_JSON','w').write(os.path.expandvars(open('$INPUT_JSON').read()))"
+
+MatPL train "$EXPANDED_JSON"

demo/nep_HfO2/dcu_1/HfO2_nep_test.json

@@ -0,0 +1,87 @@
+{
+    "model_type": "NEP",
+    "atom_type": [
+        8,
+        72
+    ],
+    "seed": 2023,
+    "recover_train": true,
+    "model": {
+        "descriptor": {
+            "cutoff": [
+                6.0,
+                6.0
+            ],
+            "n_max": [
+                4,
+                4
+            ],
+            "basis_size": [
+                12,
+                12
+            ],
+            "l_max": [
+                4,
+                2,
+                1
+            ]
+        },
+        "fitting_net": {
+            "network_size": [
+                40,
+                1
+            ]
+        }
+    },
+    "optimizer": {
+        "optimizer": "ADAM",
+        "epochs": 30,
+        "batch_size": 1,
+        "print_freq": 10,
+        "learning_rate": 0.001,
+        "stop_lr": 3.51e-08,
+        "stop_step": 1000000,
+        "decay_step": 5000,
+        "train_energy": true,
+        "train_force": true,
+        "train_virial": false,
+        "start_pre_fac_force": 1000,
+        "start_pre_fac_etot": 0.02,
+        "start_pre_fac_virial": 50.0,
+        "end_pre_fac_force": 1.0,
+        "end_pre_fac_etot": 1.0,
+        "end_pre_fac_virial": 1.0
+    },
+    "format": "pwmlff/npy",
+    "model_load_file": "./model_record/nep_model.ckpt",
+    "test_data": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_000_50",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_004_50",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_008_50",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_012_50",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_016_50",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_020_20",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_024_20",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_001_50",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_005_50",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_009_50",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_013_50",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_017_50",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_021_20",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_025_20",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_002_50",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_006_50",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_010_50",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_014_50",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_018_50",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_022_20",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_026_20",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_003_50",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_007_50",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_011_50",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_015_30",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_019_50",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_023_20",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_027_20"
+    ]
+}

demo/nep_HfO2/dcu_1/HfO2_nep_train.json

@@ -0,0 +1,91 @@
+{
+    "model_type": "NEP",
+    "atom_type": [
+        8,
+        72
+    ],
+    "seed": 2023,
+    "recover_train": true,
+    "model": {
+        "descriptor": {
+            "cutoff": [
+                6.0,
+                6.0
+            ],
+            "n_max": [
+                4,
+                4
+            ],
+            "basis_size": [
+                12,
+                12
+            ],
+            "l_max": [
+                4,
+                2,
+                1
+            ]
+        },
+        "fitting_net": {
+            "network_size": [
+                40,
+                1
+            ]
+        }
+    },
+    "optimizer": {
+        "optimizer": "ADAM",
+        "epochs": 30,
+        "batch_size": 1,
+        "print_freq": 10,
+        "learning_rate": 0.001,
+        "stop_lr": 3.51e-08,
+        "stop_step": 1000000,
+        "decay_step": 5000,
+        "train_energy": true,
+        "train_force": true,
+        "train_virial": false,
+        "start_pre_fac_force": 1000,
+        "start_pre_fac_etot": 0.02,
+        "start_pre_fac_virial": 50.0,
+        "end_pre_fac_force": 1.0,
+        "end_pre_fac_etot": 1.0,
+        "end_pre_fac_virial": 1.0
+    },
+    "format": "pwmlff/npy",
+    "train_data": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_000_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_002_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_004_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_006_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_008_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_010_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_012_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_014_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_016_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_018_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_020_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_022_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_024_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_026_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_001_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_003_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_005_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_007_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_009_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_011_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_013_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_015_30/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_017_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_019_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_021_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_023_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_025_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_027_20/"
+    ],
+    "valid_data": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_000_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_004_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_008_50/"
+    ]
+}

demo/nep_HfO2/dcu_1/submit.sh

@@ -0,0 +1,48 @@
+#!/bin/bash
+#SBATCH --job-name=nep_HfO2_train
+#SBATCH --partition=hx1hdexclu12
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=1
+#SBATCH --gres=dcu:1
+#SBATCH --cpus-per-task=16
+#SBATCH --time=2:00:00
+#SBATCH --output=slurm_%j.out
+#SBATCH --error=slurm_%j.err
+
+if [[ -n "${SCRIPT_DIR}" ]]; then
+    :
+elif [[ -n "${SLURM_SUBMIT_DIR}" ]]; then
+    SCRIPT_DIR="${SLURM_SUBMIT_DIR}"
+else
+    echo "ERROR: 未检测到外部 SCRIPT_DIR 变量，也不在 Slurm 任务环境（SLURM_SUBMIT_DIR 为空）"
+    exit 1
+fi
+echo $SCRIPT_DIR
+
+export MATCHEM_CONDA_NAME="${MATCHEM_CONDA_NAME:-test_pip}"
+source "$SCRIPT_DIR/matchem_env.sh"
+
+# MatPL 运行时环境（不侵入 matchem_env.sh，各算例自行维护）
+# 校验MatPL根路径
+if [[ -z "${MATPL_SRC_DIR}" || "${MATPL_SRC_DIR}" == "/path/to/matpl_dcu" || ! -d "${MATPL_SRC_DIR}" ]]; then
+    echo "=============================================="
+    echo "ERROR: MATPL_SRC_DIR 路径配置错误！"
+    echo "当前值：${MATPL_SRC_DIR}"
+    echo "请设置为真实有效的MatPL源码目录，示例："
+    echo "export MATPL_SRC_DIR=/public/home/xxx/real_matpl_path"
+    echo "=============================================="
+    exit 1
+fi
+
+if [ -f "$MATPL_SRC_DIR/env.sh" ]; then
+    source "$MATPL_SRC_DIR/env.sh"
+fi
+echo "MATPL_SRC_DIR = $MATPL_SRC_DIR"
+export LD_LIBRARY_PATH="$MATPL_SRC_DIR/src/op/build/lib:${LD_LIBRARY_PATH:-}"
+# 输入 JSON 使用环境变量，运行前展开为临时文件
+INPUT_JSON="HfO2_nep_train.json"
+EXPANDED_JSON=".${INPUT_JSON%.json}_expanded_$$.json"
+trap 'rm -f "$EXPANDED_JSON"' EXIT
+python3 -c "import os; open('$EXPANDED_JSON','w').write(os.path.expandvars(open('$INPUT_JSON').read()))"
+
+MatPL train "$EXPANDED_JSON"

demo/nep_HfO2/dcu_16/HfO2_nep_train_16card.json

@@ -0,0 +1,91 @@
+{
+    "model_type": "NEP",
+    "atom_type": [
+        8,
+        72
+    ],
+    "seed": 2023,
+    "recover_train": true,
+    "model": {
+        "descriptor": {
+            "cutoff": [
+                6.0,
+                6.0
+            ],
+            "n_max": [
+                4,
+                4
+            ],
+            "basis_size": [
+                12,
+                12
+            ],
+            "l_max": [
+                4,
+                2,
+                1
+            ]
+        },
+        "fitting_net": {
+            "network_size": [
+                40,
+                1
+            ]
+        }
+    },
+    "optimizer": {
+        "optimizer": "ADAM",
+        "epochs": 30,
+        "batch_size": 16,
+        "print_freq": 10,
+        "learning_rate": 0.001,
+        "stop_lr": 3.51e-08,
+        "stop_step": 1000000,
+        "decay_step": 5000,
+        "train_energy": true,
+        "train_force": true,
+        "train_virial": false,
+        "start_pre_fac_force": 1000,
+        "start_pre_fac_etot": 0.02,
+        "start_pre_fac_virial": 50.0,
+        "end_pre_fac_force": 1.0,
+        "end_pre_fac_etot": 1.0,
+        "end_pre_fac_virial": 1.0
+    },
+    "format": "pwmlff/npy",
+    "train_data": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_000_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_002_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_004_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_006_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_008_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_010_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_012_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_014_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_016_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_018_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_020_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_022_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_024_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_026_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_001_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_003_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_005_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_007_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_009_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_011_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_013_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_015_30/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_017_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_019_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_021_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_023_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_025_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_027_20/"
+    ],
+    "valid_data": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_000_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_004_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_008_50/"
+    ]
+}

demo/nep_HfO2/dcu_16/submit_16card.sh

@@ -0,0 +1,61 @@
+#!/bin/bash
+#SBATCH --job-name=nep_HfO2_16card
+#SBATCH --partition=hx1hdexclu12
+#SBATCH --nodes=2
+#SBATCH --ntasks-per-node=8
+#SBATCH --gres=dcu:8
+#SBATCH --cpus-per-task=4
+#SBATCH --mem=256G
+#SBATCH --time=2:00:00
+#SBATCH --output=slurm_16card_%j.out
+#SBATCH --error=slurm_16card_%j.err
+
+if [[ -n "${SCRIPT_DIR}" ]]; then
+    :
+elif [[ -n "${SLURM_SUBMIT_DIR}" ]]; then
+    SCRIPT_DIR="${SLURM_SUBMIT_DIR}"
+else
+    echo "ERROR: 未检测到外部 SCRIPT_DIR 变量，也不在 Slurm 任务环境（SLURM_SUBMIT_DIR 为空）"
+    exit 1
+fi
+echo $SCRIPT_DIR
+
+export MATCHEM_CONDA_NAME="${MATCHEM_CONDA_NAME:-test_pip}"
+source "$SCRIPT_DIR/matchem_env.sh"
+
+# MatPL 运行时环境（不侵入 matchem_env.sh，各算例自行维护）
+# 校验MatPL根路径
+if [[ -z "${MATPL_SRC_DIR}" || "${MATPL_SRC_DIR}" == "/path/to/matpl_dcu" || ! -d "${MATPL_SRC_DIR}" ]]; then
+    echo "=============================================="
+    echo "ERROR: MATPL_SRC_DIR 路径配置错误！"
+    echo "当前值：${MATPL_SRC_DIR}"
+    echo "请设置为真实有效的MatPL源码目录，示例："
+    echo "export MATPL_SRC_DIR=/public/home/xxx/real_matpl_path"
+    echo "=============================================="
+    exit 1
+fi
+
+if [ -f "$MATPL_SRC_DIR/env.sh" ]; then
+    source "$MATPL_SRC_DIR/env.sh"
+fi
+echo "MATPL_SRC_DIR = $MATPL_SRC_DIR"
+export LD_LIBRARY_PATH="$MATPL_SRC_DIR/src/op/build/lib:${LD_LIBRARY_PATH:-}"
+
+# 多节点通信设置
+MASTER_ADDR=$(scontrol show hostnames $SLURM_JOB_NODELIST | head -n 1)
+function get_free_port() {
+    python -c 'import socket; s = socket.socket(socket.AF_INET, socket.SOCK_STREAM); s.bind(("", 0)); print(s.getsockname()[1]); s.close()'
+}
+MASTER_PORT=$(get_free_port)
+export MASTER_ADDR=$MASTER_ADDR
+export MASTER_PORT=$MASTER_PORT
+
+# 16 卡训练（2 节点 × 8 DCU）
+
+# 输入 JSON 使用环境变量，运行前展开为临时文件
+INPUT_JSON="HfO2_nep_train_16card.json"
+EXPANDED_JSON=".${INPUT_JSON%.json}_expanded_$$.json"
+trap 'rm -f "$EXPANDED_JSON"' EXIT
+python3 -c "import os; open('$EXPANDED_JSON','w').write(os.path.expandvars(open('$INPUT_JSON').read()))"
+
+srun MatPL train "$EXPANDED_JSON"

demo/nep_HfO2/dcu_8/HfO2_nep_train_8card.json

@@ -0,0 +1,91 @@
+{
+    "model_type": "NEP",
+    "atom_type": [
+        8,
+        72
+    ],
+    "seed": 2023,
+    "recover_train": true,
+    "model": {
+        "descriptor": {
+            "cutoff": [
+                6.0,
+                6.0
+            ],
+            "n_max": [
+                4,
+                4
+            ],
+            "basis_size": [
+                12,
+                12
+            ],
+            "l_max": [
+                4,
+                2,
+                1
+            ]
+        },
+        "fitting_net": {
+            "network_size": [
+                40,
+                1
+            ]
+        }
+    },
+    "optimizer": {
+        "optimizer": "ADAM",
+        "epochs": 30,
+        "batch_size": 8,
+        "print_freq": 10,
+        "learning_rate": 0.001,
+        "stop_lr": 3.51e-08,
+        "stop_step": 1000000,
+        "decay_step": 5000,
+        "train_energy": true,
+        "train_force": true,
+        "train_virial": false,
+        "start_pre_fac_force": 1000,
+        "start_pre_fac_etot": 0.02,
+        "start_pre_fac_virial": 50.0,
+        "end_pre_fac_force": 1.0,
+        "end_pre_fac_etot": 1.0,
+        "end_pre_fac_virial": 1.0
+    },
+    "format": "pwmlff/npy",
+    "train_data": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_000_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_002_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_004_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_006_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_008_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_010_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_012_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_014_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_016_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_018_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_020_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_022_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_024_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_026_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_001_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_003_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_005_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_007_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_009_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_011_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_013_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_015_30/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_017_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_019_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_021_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_023_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_025_20/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_027_20/"
+    ],
+    "valid_data": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_000_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_004_50/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/HfO2/pwdata/init_008_50/"
+    ]
+}

demo/nep_HfO2/dcu_8/submit_8card.sh

@@ -0,0 +1,50 @@
+#!/bin/bash
+#SBATCH --job-name=nep_HfO2_8card
+#SBATCH --partition=hx1hdexclu12
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=8
+#SBATCH --gres=dcu:8
+#SBATCH --cpus-per-task=4
+#SBATCH --mem=400G
+#SBATCH --time=2:00:00
+#SBATCH --output=slurm_8card_%j.out
+#SBATCH --error=slurm_8card_%j.err
+
+if [[ -n "${SCRIPT_DIR}" ]]; then
+    :
+elif [[ -n "${SLURM_SUBMIT_DIR}" ]]; then
+    SCRIPT_DIR="${SLURM_SUBMIT_DIR}"
+else
+    echo "ERROR: 未检测到外部 SCRIPT_DIR 变量，也不在 Slurm 任务环境（SLURM_SUBMIT_DIR 为空）"
+    exit 1
+fi
+echo $SCRIPT_DIR
+
+export MATCHEM_CONDA_NAME="${MATCHEM_CONDA_NAME:-test_pip}"
+source "$SCRIPT_DIR/matchem_env.sh"
+
+# MatPL 运行时环境（不侵入 matchem_env.sh，各算例自行维护）
+# 校验MatPL根路径
+if [[ -z "${MATPL_SRC_DIR}" || "${MATPL_SRC_DIR}" == "/path/to/matpl_dcu" || ! -d "${MATPL_SRC_DIR}" ]]; then
+    echo "=============================================="
+    echo "ERROR: MATPL_SRC_DIR 路径配置错误！"
+    echo "当前值：${MATPL_SRC_DIR}"
+    echo "请设置为真实有效的MatPL源码目录，示例："
+    echo "export MATPL_SRC_DIR=/public/home/xxx/real_matpl_path"
+    echo "=============================================="
+    exit 1
+fi
+
+if [ -f "$MATPL_SRC_DIR/env.sh" ]; then
+    source "$MATPL_SRC_DIR/env.sh"
+fi
+echo "MATPL_SRC_DIR = $MATPL_SRC_DIR"
+export LD_LIBRARY_PATH="$MATPL_SRC_DIR/src/op/build/lib:${LD_LIBRARY_PATH:-}"
+
+# 输入 JSON 使用环境变量，运行前展开为临时文件
+INPUT_JSON="HfO2_nep_train_8card.json"
+EXPANDED_JSON=".${INPUT_JSON%.json}_expanded_$$.json"
+trap 'rm -f "$EXPANDED_JSON"' EXIT
+python3 -c "import os; open('$EXPANDED_JSON','w').write(os.path.expandvars(open('$INPUT_JSON').read()))"
+
+MatPL train "$EXPANDED_JSON"

demo/nep_LiSiC/LiSiC_nep_train.json

@@ -0,0 +1,72 @@
+{
+    "model_type": "NEP",
+    "atom_type": [
+        3,
+        14,
+        6
+    ],
+    "seed": 2023,
+    "recover_train": true,
+    "model": {
+        "descriptor": {
+            "cutoff": [
+                6.0,
+                6.0
+            ],
+            "n_max": [
+                4,
+                4
+            ],
+            "basis_size": [
+                12,
+                12
+            ],
+            "l_max": [
+                4,
+                2,
+                1
+            ]
+        },
+        "fitting_net": {
+            "network_size": [
+                40,
+                1
+            ]
+        }
+    },
+    "optimizer": {
+        "optimizer": "ADAM",
+        "epochs": 30,
+        "batch_size": 1,
+        "print_freq": 10,
+        "learning_rate": 0.001,
+        "stop_lr": 3.51e-08,
+        "stop_step": 1000000,
+        "decay_step": 5000,
+        "train_energy": true,
+        "train_force": true,
+        "train_virial": false,
+        "start_pre_fac_force": 1000,
+        "start_pre_fac_etot": 0.02,
+        "start_pre_fac_virial": 50.0,
+        "end_pre_fac_force": 1.0,
+        "end_pre_fac_etot": 1.0,
+        "end_pre_fac_virial": 1.0
+    },
+    "format": "pwmlff/npy",
+    "train_data": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/C2",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/C448",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/C448Li75",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/C64Si32",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/Li1Si24",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/Li3Si8",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/Li8",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/Li88Si20",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/Si1",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/Si217"
+    ],
+    "valid_data": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/C2"
+    ]
+}

demo/nep_LiSiC/std_input.json

@@ -0,0 +1,72 @@
+{
+    "model_type": "NEP",
+    "atom_type": [
+        3,
+        14,
+        6
+    ],
+    "seed": 2023,
+    "recover_train": true,
+    "model": {
+        "descriptor": {
+            "cutoff": [
+                6.0,
+                6.0
+            ],
+            "n_max": [
+                4,
+                4
+            ],
+            "basis_size": [
+                12,
+                12
+            ],
+            "l_max": [
+                4,
+                2,
+                1
+            ]
+        },
+        "fitting_net": {
+            "network_size": [
+                40,
+                1
+            ]
+        }
+    },
+    "optimizer": {
+        "optimizer": "ADAM",
+        "epochs": 30,
+        "batch_size": 1,
+        "print_freq": 10,
+        "learning_rate": 0.001,
+        "stop_lr": 3.51e-08,
+        "stop_step": 1000000,
+        "decay_step": 5000,
+        "train_energy": true,
+        "train_force": true,
+        "train_virial": false,
+        "start_pre_fac_force": 1000,
+        "start_pre_fac_etot": 0.02,
+        "start_pre_fac_virial": 50.0,
+        "end_pre_fac_force": 1.0,
+        "end_pre_fac_etot": 1.0,
+        "end_pre_fac_virial": 1.0
+    },
+    "format": "pwmlff/npy",
+    "train_data": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/C2",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/C448",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/C448Li75",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/C64Si32",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/Li1Si24",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/Li3Si8",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/Li8",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/Li88Si20",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/Si1",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/Si217"
+    ],
+    "valid_data": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/matpl/LiSiC/pwdata/C2"
+    ]
+}

demo/nep_LiSiC/submit.sh

@@ -0,0 +1,51 @@
+#!/bin/bash
+#SBATCH --job-name=nep_LiSiC_train
+#SBATCH --partition=hx1hdexclu12
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=1
+#SBATCH --gres=dcu:1
+#SBATCH --cpus-per-task=16
+#SBATCH --time=2:00:00
+#SBATCH --output=slurm_%j.out
+#SBATCH --error=slurm_%j.err
+
+if [[ -n "${SCRIPT_DIR}" ]]; then
+    :
+elif [[ -n "${SLURM_SUBMIT_DIR}" ]]; then
+    SCRIPT_DIR="${SLURM_SUBMIT_DIR}"
+else
+    echo "ERROR: 未检测到外部 SCRIPT_DIR 变量，也不在 Slurm 任务环境（SLURM_SUBMIT_DIR 为空）"
+    exit 1
+fi
+echo $SCRIPT_DIR
+
+export MATCHEM_CONDA_NAME="${MATCHEM_CONDA_NAME:-test_pip}"
+source "$SCRIPT_DIR/matchem_env.sh"
+
+# MatPL 运行时环境（不侵入 matchem_env.sh，各算例自行维护）
+# 校验MatPL根路径
+if [[ -z "${MATPL_SRC_DIR}" || "${MATPL_SRC_DIR}" == "/path/to/matpl_dcu" || ! -d "${MATPL_SRC_DIR}" ]]; then
+    echo "=============================================="
+    echo "ERROR: MATPL_SRC_DIR 路径配置错误！"
+    echo "当前值：${MATPL_SRC_DIR}"
+    echo "请设置为真实有效的MatPL源码目录，示例："
+    echo "export MATPL_SRC_DIR=/public/home/xxx/real_matpl_path"
+    echo "=============================================="
+    exit 1
+fi
+
+if [ -f "$MATPL_SRC_DIR/env.sh" ]; then
+    source "$MATPL_SRC_DIR/env.sh"
+fi
+echo "MATPL_SRC_DIR = $MATPL_SRC_DIR"
+export LD_LIBRARY_PATH="$MATPL_SRC_DIR/src/op/build/lib:${LD_LIBRARY_PATH:-}"
+
+# 单卡训练
+
+# 输入 JSON 使用环境变量，运行前展开为临时文件
+INPUT_JSON="LiSiC_nep_train.json"
+EXPANDED_JSON=".${INPUT_JSON%.json}_expanded_$$.json"
+trap 'rm -f "$EXPANDED_JSON"' EXIT
+python3 -c "import os; open('$EXPANDED_JSON','w').write(os.path.expandvars(open('$INPUT_JSON').read()))"
+
+MatPL train "$EXPANDED_JSON"

matchem_env.sh

@@ -0,0 +1,48 @@
+#!/bin/bash
+# ==========================================
+# MatChem 统一环境配置脚本
+# 用途：加载模块、激活 conda、导出各组件路径
+# 用法：source matchem_env.sh
+# 注意：默认激活 conda 环境 test_pip；若使用其他环境名，请执行：
+#       MATCHEM_CONDA_NAME=your_env source matchem_env.sh
+# ==========================================
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+
+# ---------- 1. 基础环境配置 ----------
+export MATCHEM_CONDA_NAME="${MATCHEM_CONDA_NAME:-test_pip}"
+
+# ---------- 2. OneScience 运行时环境变量 ----------
+export ONESCIENCE_DATASETS_DIR="/public/share/sugonhpcapp01/onestore/onedatasets"
+export ONESCIENCE_MODELS_DIR="/public/share/sugonhpcapp01/onestore/onemodels"
+export device="gpu"  # 根据实际平台改为 gpu 或 dcu
+export LD_LIBRARY_PATH="${CONDA_PREFIX:-}/lib:${LD_LIBRARY_PATH:-}"
+
+# ---------- 3. 外部软件路径（使用 dp_install.sh / matpl_install.sh / lmp_install.sh 时会自动更新） ----------
+# DeepMD-kit 源码目录
+export DEEPMD_SRC_DIR="${SCRIPT_DIR}/deepmd-kit"
+
+# MatPL 源码目录
+export MATPL_SRC_DIR="${SCRIPT_DIR}/matpl_dcu"
+
+# LAMMPS 安装目录
+export LAMMPS_INSTALL_DIR="${SCRIPT_DIR}/lammps_dcu"
+
+# DeepMD C++ 接口目录
+export DP_CPP_DIR="${SCRIPT_DIR}/dp_cpp_dcu"
+
+# ---------- 4. 加载集群模块与 conda ----------
+set +u
+source ~/.bashrc
+set -u
+module load sghpcdas/25.6        # DTK / PyTorch 等 SDK
+module load sghpc-mpi-gcc/26.3   # MPI 与 GCC 编译器
+
+conda activate "$MATCHEM_CONDA_NAME"
+
+# ---------- 6. LAMMPS 运行时环境 ----------
+export LD_LIBRARY_PATH=${LAMMPS_INSTALL_DIR}/lib64:${LD_LIBRARY_PATH:-}
+export LD_LIBRARY_PATH=${LAMMPS_INSTALL_DIR}/lib_override:${LD_LIBRARY_PATH:-}
+export LAMMPS_PLUGIN_PATH=${DP_CPP_DIR}/lib
+
+echo "✅ MatChem 环境已激活: ${MATCHEM_CONDA_NAME}"

matpl_install.sh

@@ -0,0 +1,179 @@
+#!/bin/bash
+# -----------------------------------------------------------------------------
+# MatPL DCU 一键安装脚本
+# 流程：源码拉取 → 编译安装 → 安装验证
+# 用法：bash matpl_install.sh
+#       MATPL_SRC_DIR=/path/to/src bash matpl_install.sh  # 指定源码路径
+# -----------------------------------------------------------------------------
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+cd "$SCRIPT_DIR"
+
+# 1. 环境准备
+echo ">>> Step 1: 加载环境"
+source "$SCRIPT_DIR/../matchem_env.sh"
+
+# 1.5 确保 gflags/glog 运行库存在（torch cmake 的运行时依赖，pip 环境通常缺失）
+echo ">>> Step 1.5: 检查 gflags/glog 运行库"
+MISSING_PKGS=""
+ls "$CONDA_PREFIX"/lib/libgflags.so* >/dev/null 2>&1 || MISSING_PKGS="${MISSING_PKGS} gflags"
+ls "$CONDA_PREFIX"/lib/libglog.so* >/dev/null 2>&1 || MISSING_PKGS="${MISSING_PKGS} glog"
+if [ -n "${MISSING_PKGS}" ]; then
+    echo ">>> 安装缺失的运行库:${MISSING_PKGS}（conda-forge）"
+    conda install -y -c conda-forge ${MISSING_PKGS}
+else
+    echo ">>> gflags/glog 已存在，跳过"
+fi
+
+# 下载辅助：优先 curl，失败时回退 wget（部分节点 curl 存在 TLS/代理问题）
+download_file() {
+    local url="$1" out="$2"
+    if command -v curl >/dev/null 2>&1 && curl -fL -o "$out" "$url"; then
+        return 0
+    fi
+    echo "[提示] curl 下载失败，改用 wget: $url"
+    if command -v wget >/dev/null 2>&1 && wget -O "$out" "$url"; then
+        return 0
+    fi
+    echo "[错误] 下载失败: $url"
+    return 1
+}
+
+# 2. 交互式配置源码路径
+MATPL_SRC="${MATPL_SRC_DIR:-${SCRIPT_DIR}/matpl_dcu}"
+if [ -t 0 ]; then
+    read -rp "请输入 MatPL 源码路径 [默认: ${MATPL_SRC}]: " input_src
+    MATPL_SRC="${input_src:-${MATPL_SRC}}"
+fi
+echo "[提示] 使用 MatPL 源码路径: ${MATPL_SRC}"
+
+# 3. 源码准备
+# 说明：
+#   - 开发/测试阶段：自动通过 HTTPS + 代理拉取源码
+#   - 生产/客户场景：建议提前上传源码到集群，通过 MATPL_SRC_DIR 指定
+if [ ! -d "$MATPL_SRC/.git" ] && [ ! -f "$MATPL_SRC/main.py" ]; then
+    echo ">>> Step 3: 拉取 MatPL DCU 源码"
+    # 计算节点需通过 HTTP 代理访问外网，配置 git 代理，如代理实效，可向集群管理员重新申请
+    git config --global http.proxy "http://scnethpc2601:sWMtqVS@10.16.1.52:3120"
+    git clone --depth 1 --branch nep-dcu/2026.3 "https://gitee.com/wang-rui-sugon/matpl_dcu.git" "$MATPL_SRC"
+else
+    echo ">>> Step 3: 源码已存在，跳过拉取"
+fi
+
+cd "$MATPL_SRC"
+
+echo "=========================================="
+echo " MatPL DCU 一键安装脚本"
+echo "=========================================="
+echo "工作目录: $MATPL_SRC"
+
+# 3.5 Patch MatPL CMakeLists.txt：避免在登录节点触发 torch 动态库加载
+echo ">>> Step 3.5: Patch MatPL CMakeLists.txt，避免登录节点 import torch"
+OP_CMAKE="$MATPL_SRC/src/op/CMakeLists.txt"
+CHEB_CMAKE="$MATPL_SRC/src/feature/chebyshev/CMakeLists.txt"
+
+if [ -f "$OP_CMAKE" ]; then
+    sed -i "s|import torch; print(torch.utils.cmake_prefix_path)|import importlib.util, os; spec = importlib.util.find_spec('torch'); p = os.path.dirname(spec.origin) if spec and spec.origin else ''; print((p + '/share/cmake') if p else '')|" "$OP_CMAKE"
+    sed -i "s|import torch; print(torch.__path__\[0\])|import importlib.util, os; spec = importlib.util.find_spec('torch'); print(os.path.dirname(spec.origin) if spec and spec.origin else '')|" "$OP_CMAKE"
+    sed -i "s|import torch; print(torch.version.hip is not None)|print(True)|" "$OP_CMAKE"
+    echo ">>> 已 patch $OP_CMAKE"
+fi
+
+if [ -f "$CHEB_CMAKE" ]; then
+    sed -i "s|import torch; print(torch.utils.cmake_prefix_path)|import importlib.util, os; spec = importlib.util.find_spec('torch'); p = os.path.dirname(spec.origin) if spec and spec.origin else ''; print((p + '/share/cmake') if p else '')|" "$CHEB_CMAKE"
+    echo ">>> 已 patch $CHEB_CMAKE"
+fi
+
+# 4. 修复 torch cmake 中硬编码的 /opt/dtk 路径
+TORCH_CMAKE="$CONDA_PREFIX/lib/python3.11/site-packages/torch/share/cmake/Caffe2/Caffe2Targets.cmake"
+if [ -f "$TORCH_CMAKE" ]; then
+    if grep -q "/opt/dtk" "$TORCH_CMAKE"; then
+        echo ">>> Step 4: 修复 torch cmake 硬编码路径"
+        sed -i "s|/opt/dtk|$ROCM_PATH|g" "$TORCH_CMAKE"
+    else
+        echo ">>> Step 4: torch cmake 路径已正确，跳过修复"
+    fi
+else
+    echo "警告: 未找到 torch Caffe2Targets.cmake，跳过修复"
+fi
+
+# 5. 确保安装兼容版 glog（0.7+ API 不兼容）
+echo ">>> Step 5: 确保 glog 0.6 已安装"
+if [ -f "$CONDA_PREFIX/lib/libglog.so.1" ]; then
+    echo "glog 0.6+ 已存在，跳过安装"
+else
+    echo "从源码编译 glog 0.6 ..."
+    GLOG_BUILD_DIR=$(mktemp -d)
+    cd "$GLOG_BUILD_DIR"
+    download_file "https://github.com/google/glog/archive/refs/tags/v0.6.0.tar.gz" v0.6.0.tar.gz || exit 1
+    tar -xzf v0.6.0.tar.gz
+    cd glog-0.6.0
+    cmake -S . -B build \
+        -DCMAKE_INSTALL_PREFIX="$CONDA_PREFIX" \
+        -DBUILD_SHARED_LIBS=ON \
+        -DWITH_GTEST=OFF
+    cmake --build build -j$(nproc)
+    cmake --install build
+    cd "$MATPL_SRC"
+    rm -rf "$GLOG_BUILD_DIR"
+fi
+
+# 6. 导出编译所需环境变量
+echo ">>> Step 6: 导出编译环境变量"
+export ROCM_PATH="$ROCM_PATH"
+export CUDA_TOOLKIT_ROOT_DIR="${ROCM_PATH}/cuda/cuda-12"
+export PATH="${ROCM_PATH}/cuda/cuda-12/bin:$PATH"
+export CMAKE_PREFIX_PATH="${ROCM_PATH}/lib/cmake:${ROCM_PATH}/dcc/comgr/lib64/cmake/amd_comgr:${CMAKE_PREFIX_PATH:-}"
+export LIBRARY_PATH="${ROCM_PATH}/cuda/cuda-12/targets/x86_64-linux/lib:${ROCM_PATH}/dcc/lib/clang/17.0.0/lib/linux:${LIBRARY_PATH:-}"
+export CPLUS_INCLUDE_PATH="$CONDA_PREFIX/include:${CPLUS_INCLUDE_PATH:-}"
+
+# 修复 GCC 版本：强制使用 sghpc-mpi-gcc/26.3 提供的 GCC 12.4.0
+export CC=/public/software/sghpc_sdk.bak/Linux_x86_64/26.3/compilers/gcc-12.4.0/bin/gcc
+export CXX=/public/software/sghpc_sdk.bak/Linux_x86_64/26.3/compilers/gcc-12.4.0/bin/g++
+
+# 7. 清理历史编译产物并重新编译
+echo ">>> Step 7: 开始编译 MatPL"
+cd src
+rm -rf feature/nep_find_neigh/build feature/NEP_GPU/build op/build
+
+# 登录节点内存有限，限制并行编译数以避免 OOM
+NPROC=$(nproc)
+if [ "$NPROC" -gt 2 ]; then
+    NPROC=2
+fi
+echo ">>> 限制并行编译数为 ${NPROC}（避免节点 OOM）"
+bash build.sh -j${NPROC}
+
+# 8. 生成 env.sh
+cd "$MATPL_SRC"
+cat > env.sh <<EOF
+# Load for MatPL
+export PYTHONPATH=$MATPL_SRC:\${PYTHONPATH:-}
+export PATH=$MATPL_SRC/src/bin:\${PATH:-}
+EOF
+
+# 9. 验证
+echo ">>> Step 9: 验证安装"
+source env.sh
+python -c "import matpl; print('MatPL import OK')" 2>/dev/null || \
+    python -c "import sys; sys.path.insert(0, '$MATPL_SRC'); import matpl; print('MatPL import OK')" 2>/dev/null || \
+    echo "注意: MatPL Python 包导入验证跳过（不影响 C++ 算子使用）"
+
+# 10. 将配置写回 matchem_env.sh
+MATCHEM_ENV_FILE="${SCRIPT_DIR}/../matchem_env.sh"
+if [ -f "${MATCHEM_ENV_FILE}" ]; then
+    echo "[提示] 更新 ${MATCHEM_ENV_FILE} ..."
+    sed -i "s|^export MATPL_SRC_DIR=.*|export MATPL_SRC_DIR=${MATPL_SRC}|" "${MATCHEM_ENV_FILE}"
+else
+    echo "[警告] 未找到 ${MATCHEM_ENV_FILE}，跳过写入配置。"
+fi
+
+echo ""
+echo "=========================================="
+echo " MatPL DCU 安装完成!"
+echo "=========================================="
+echo "源码路径: $MATPL_SRC"
+echo "环境文件: $MATPL_SRC/env.sh"
+echo ""
+echo "每次使用前请执行:"
+echo "  source $SCRIPT_DIR/../matchem_env.sh"