Upload folder using huggingface_hub

README.md

@@ -0,0 +1,119 @@
+<p align="center">
+  <strong>
+    <span style="font-size: 30px;">DeePMD</span>
+  </strong>
+</p>
+
+# Model Introduction
+
+DeePMD is a Deep Potential Molecular Dynamics model ecosystem for machine-learning potential function training of atomic systems. It provides a minimal runnable training entry point based on PyTorch/TensorFlow backends.
+
+Reference implementation: DeepMD-kit project
+
+# Model Description
+
+DeePMD is based on a deep neural network architecture and is trained on atomic-system data. It performs interatomic potential function training and molecular dynamics simulation for molecular and materials systems.
+
+# Applicable Scenarios
+
+| Scenario | Description |
+| :---: | :--- |
+| DeePMD water training | Train the water potential using configurations such as `demo/water_se_e2_a_pt/input_torch.json` |
+| Multi-GPU SLURM submission | Refer to `demo/water_se_e2_a_pt/submit_4card.sh` and `submit_8card.sh` |
+| Environment connectivity check | Run `dp_install.sh` to check whether DeepMD-kit can be installed |
+| Custom data migration | Replace the `systems` paths in the configuration file with your own DeepMD npy data |
+
+# 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**
+
+- GPU or DCU is recommended for training.
+- CPU can be used for installation checks and small-data connectivity verification; 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/DeePMD --local_dir ./deepmd
+cd deepmd
+```
+
+### 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 DeepMD-kit
+
+```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 dp_install.sh
+```
+
+### Training Data Description
+
+This repository does not include built-in training data. Taking the DeePMD water dataset as an example, download it from ModelScope and place it under `data/` in the repository root:
+
+```bash
+modelscope download --dataset OneScience/DeePMD --local_dir ./data
+```
+
+After downloading, the data path will be `data/DeePMD/water/data_0..3/`.
+
+### Training
+
+Single-GPU:
+
+```bash
+cd demo/water_se_e2_a_pt
+dp --pt train input_torch.json
+```
+
+Multi-GPU:
+
+```bash
+cd demo/water_se_e2_a_pt
+bash submit_4card.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 DeePMD example code comes from the matchem example implementation in the OneScience project and references the upstream DeepMD-kit project. For upstream DeepMD-kit licensing information, please refer to its official repository.
+- If you use DeePMD training results in scientific research, we recommend citing DeepMD-kit, the relevant OneScience project information, and the sources of the datasets actually used.

demo/dpa3_finetune_CH/DPA-3.1-3M.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:86dd3a804d78ca5d203ebf98747e8f16dff9713ba8950097ceb760b161e19907
+size 47176032

demo/dpa3_finetune_CH/convert_ase_to_dp.py

@@ -0,0 +1,71 @@
+#!/usr/bin/env python3
+"""Convert ASE LMDB database to DeepMD-kit numpy format, grouped by natoms.
+Non-periodic molecules are placed in a cubic box large enough to avoid
+self-images across the cutoff."""
+
+import os
+import numpy as np
+from ase.db import connect
+
+TYPE_MAP = ["H", "C"]
+# Largest cutoff in the DPA-3 repflow config; add margin to avoid self-images
+MAX_CUTOFF = 6.0
+BOX_MARGIN = 2 * MAX_CUTOFF + 4.0  # ~16 Å margin on each side
+
+def make_cubic_box(coords):
+    """Build a cubic box that contains the molecule with enough margin."""
+    min_pos = coords.min(axis=0)
+    max_pos = coords.max(axis=0)
+    extent = (max_pos - min_pos).max()
+    box_len = extent + BOX_MARGIN
+    # Place molecule centered in the box
+    shift = box_len / 2.0 - (min_pos + max_pos) / 2.0
+    return np.eye(3) * box_len, shift
+
+def convert_aselmdb_to_deepmd(input_file, output_base):
+    db = connect(input_file)
+    frames_by_natoms = {}
+    for row in db.select():
+        atoms = row.toatoms()
+        natoms = len(atoms)
+        atoms_type = [TYPE_MAP.index(sym) for sym in atoms.get_chemical_symbols()]
+        coords = atoms.get_positions()
+        box, shift = make_cubic_box(coords)
+        coords = coords + shift
+        frames_by_natoms.setdefault(natoms, []).append({
+            "coords": coords,
+            "types": atoms_type,
+            "energy": atoms.get_potential_energy(),
+            "forces": atoms.get_forces(),
+            "box": box,
+        })
+
+    system_dirs = []
+    for natoms, frames in sorted(frames_by_natoms.items()):
+        nframes = len(frames)
+        system_dir = os.path.join(output_base, f"sys_{natoms}")
+        set_dir = os.path.join(system_dir, "set.000")
+        os.makedirs(set_dir, exist_ok=True)
+
+        coord = np.stack([f["coords"] for f in frames]).reshape(nframes, -1)
+        force = np.stack([f["forces"] for f in frames]).reshape(nframes, -1)
+        energy = np.array([f["energy"] for f in frames])
+        box = np.stack([f["box"] for f in frames])
+        types = np.array(frames[0]["types"])
+
+        np.save(os.path.join(set_dir, "coord.npy"), coord)
+        np.save(os.path.join(set_dir, "force.npy"), force)
+        np.save(os.path.join(set_dir, "energy.npy"), energy)
+        np.save(os.path.join(set_dir, "box.npy"), box)
+        with open(os.path.join(system_dir, "type.raw"), "w") as f:
+            f.write(" ".join(map(str, types)) + "\n")
+
+        system_dirs.append(system_dir)
+        print(f"  {system_dir}: {nframes} frames, {natoms} atoms")
+
+    print(f"Converted {input_file} -> {output_base}: {sum(len(v) for v in frames_by_natoms.values())} frames in {len(frames_by_natoms)} systems")
+    return system_dirs
+
+if __name__ == "__main__":
+    convert_aselmdb_to_deepmd("my_data_CH_3787_train.aselmdb", "train_CH")
+    convert_aselmdb_to_deepmd("my_data_CH_3787_val.aselmdb", "val_CH")

demo/dpa3_finetune_CH/input_finetune.json

@@ -0,0 +1,104 @@
+{
+  "_comment": "DPA-3 fine-tune on CH polymer data",
+  "model": {
+    "type_map": [
+      "H",
+      "C"
+    ],
+    "descriptor": {
+      "type": "dpa3",
+      "repflow": {
+        "n_dim": 128,
+        "e_dim": 64,
+        "a_dim": 32,
+        "nlayers": 16,
+        "e_rcut": 6.0,
+        "e_rcut_smth": 5.3,
+        "e_sel": 1200,
+        "a_rcut": 4.0,
+        "a_rcut_smth": 3.5,
+        "a_sel": 300,
+        "axis_neuron": 4,
+        "fix_stat_std": 0.3,
+        "a_compress_rate": 1,
+        "a_compress_e_rate": 2,
+        "a_compress_use_split": true,
+        "update_angle": true,
+        "smooth_edge_update": true,
+        "use_dynamic_sel": true,
+        "sel_reduce_factor": 10.0,
+        "use_exp_switch": true,
+        "update_style": "res_residual",
+        "update_residual": 0.1,
+        "update_residual_init": "const"
+      },
+      "activation_function": "silut:3.0",
+      "use_tebd_bias": false,
+      "precision": "float32",
+      "concat_output_tebd": false,
+      "trainable": false
+    },
+    "fitting_net": {
+      "neuron": [
+        240,
+        240,
+        240
+      ],
+      "dim_case_embd": 31,
+      "resnet_dt": true,
+      "precision": "float32",
+      "activation_function": "silut:3.0",
+      "seed": 1,
+      "_comment": " that's all"
+    },
+    "_comment": " that's all"
+  },
+  "learning_rate": {
+    "type": "exp",
+    "decay_steps": 5000,
+    "start_lr": 0.0001,
+    "stop_lr": 3e-6,
+    "_comment": "that's all"
+  },
+  "loss": {
+    "type": "ener",
+    "start_pref_e": 0.2,
+    "limit_pref_e": 20,
+    "start_pref_f": 100,
+    "limit_pref_f": 60,
+    "start_pref_v": 0.02,
+    "limit_pref_v": 1,
+    "_comment": " that's all"
+  },
+  "training": {
+    "stat_file": "./dpa3.hdf5",
+    "training_data": {
+      "systems": [
+        "$/public/share/sugonhpcapp01/onestore/onedatasets/matchem/dpa3/CH_3787/train_CH"
+      ],
+      "rglob_patterns": [
+        "sys_*"
+      ],
+      "batch_size": 1,
+      "_comment": "that's all"
+    },
+    "validation_data": {
+      "systems": [
+        "$/public/share/sugonhpcapp01/onestore/onedatasets//matchem/dpa3/CH_3787/val_CH"
+      ],
+      "rglob_patterns": [
+        "sys_*"
+      ],
+      "batch_size": 1,
+      "_comment": "that's all"
+    },
+    "numb_steps": 10000,
+    "warmup_steps": 0,
+    "gradient_max_norm": 5.0,
+    "seed": 10,
+    "disp_file": "lcurve.out",
+    "disp_freq": 1000,
+    "save_freq": 10000,
+    "_comment": "that's all"
+  }
+}

demo/dpa3_finetune_CH/submit.sh

@@ -0,0 +1,34 @@
+#!/bin/bash
+#SBATCH --job-name=dpa3_finetune
+#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="${SLURM_SUBMIT_DIR}"
+else
+    echo "ERROR: 未检测到外部 SCRIPT_DIR 变量，也不在 Slurm 任务环境（SLURM_SUBMIT_DIR 为空）"
+    exit 1
+fi
+export MATCHEM_CONDA_NAME="${MATCHEM_CONDA_NAME:-test_pip}"
+source "${SCRIPT_DIR}/matchem_env.sh"
+echo $ONESCIENCE_DATASETS_DIR
+# 限制并行度，规避 ROCm kernel launch 问题
+export DP_INTRA_OP_PARALLELISM_THREADS=1
+export DP_INTER_OP_PARALLELISM_THREADS=1
+export OMP_NUM_THREADS=1
+
+# 输入 JSON 使用环境变量，运行前展开为临时文件
+INPUT_JSON="input_finetune.json"
+EXPANDED_JSON=".${INPUT_JSON%.json}_expanded_$$.json"
+trap 'rm -f "$EXPANDED_JSON"' EXIT
+python3 -c "import os; open('$EXPANDED_JSON','w').write(os.path.expandvars(open('$INPUT_JSON').read()))"
+
+dp --pt train "$EXPANDED_JSON" --finetune ./DPA-3.1-3M.pt

demo/water_se_atten_pt/input_torch.json

@@ -0,0 +1,84 @@
+{
+  "model": {
+    "type_map": [
+      "O",
+      "H"
+    ],
+    "descriptor": {
+      "type": "dpa1",
+      "sel": 120,
+      "rcut_smth": 0.5,
+      "rcut": 6.0,
+      "neuron": [
+        25,
+        50,
+        100
+      ],
+      "tebd_dim": 8,
+      "axis_neuron": 16,
+      "type_one_side": true,
+      "attn": 128,
+      "attn_layer": 2,
+      "attn_dotr": true,
+      "attn_mask": false,
+      "activation_function": "tanh",
+      "scaling_factor": 1.0,
+      "normalize": true,
+      "temperature": 1.0
+    },
+    "fitting_net": {
+      "neuron": [
+        240,
+        240,
+        240
+      ],
+      "resnet_dt": true,
+      "seed": 1,
+      "_comment": " that's all"
+    },
+    "data_stat_nbatch": 20,
+    "_comment": " that's all"
+  },
+  "learning_rate": {
+    "type": "exp",
+    "decay_steps": 5000,
+    "start_lr": 0.001,
+    "stop_lr": 3.51e-8,
+    "_comment": "that's all"
+  },
+  "loss": {
+    "type": "ener",
+    "start_pref_e": 0.02,
+    "limit_pref_e": 1,
+    "start_pref_f": 1000,
+    "limit_pref_f": 1,
+    "_comment": " that's all"
+  },
+  "training": {
+    "stat_file": "./dpa1.hdf5",
+    "training_data": {
+      "systems": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/dp/water/data_0",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/dp/water/data_1",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/dp/water/data_2"
+      ],
+      "batch_size": 4,
+      "_comment": "that's all"
+    },
+    "validation_data": {
+      "systems": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/dp/water/data_3"
+      ],
+      "batch_size": 4,
+      "numb_btch": 3,
+      "_comment": "that's all"
+    },
+    "numb_steps": 1000,
+    "seed": 10,
+    "disp_file": "lcurve.out",
+    "disp_freq": 100,
+    "save_freq": 1000,
+    "_comment": "that's all"
+  },
+  "_comment": "that's all"
+}

demo/water_se_atten_pt/submit_1card.sh

@@ -0,0 +1,34 @@
+#!/bin/bash
+#SBATCH --job-name=dp_pt_atten
+#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 $SCRIPT_DIR
+
+source "$SCRIPT_DIR/matchem_env.sh"
+
+# DeepMD 训练环境已由 matchem_env.sh 覆盖
+
+# 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="input_torch.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('$INPUT_JSON').read()))"
+
+dp --pt train "$EXPANDED_JSON"

demo/water_se_atten_pt/submit_4card.sh

@@ -0,0 +1,35 @@
+#!/bin/bash
+#SBATCH --job-name=dp_pt_atten_4card
+#SBATCH --partition=hx1hdexclu12
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=1
+#SBATCH --gres=dcu:4
+#SBATCH --cpus-per-task=16
+#SBATCH --time=2:00:00
+#SBATCH --output=slurm_4card_%j.out
+#SBATCH --error=slurm_4card_%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 $SCRIPT_DIR
+
+source "$SCRIPT_DIR/matchem_env.sh"
+
+# DeepMD 训练环境已由 matchem_env.sh 覆盖
+
+# 多卡训练（4卡）
+cd "$SCRIPT_DIR"
+
+# 输入 JSON 使用环境变量，运行前展开为临时文件
+INPUT_JSON="input_torch.json"
+EXPANDED_JSON=".${INPUT_JSON%.json}_expanded_$$.json"
+trap 'rm -f "$EXPANDED_JSON"' EXIT
+python3 -c "import os; open('$EXPANDED_JSON','w').write(os.path.expandvars(open('$INPUT_JSON').read()))"
+
+torchrun --nproc_per_node=4 -m deepmd --pt train "$EXPANDED_JSON"

demo/water_se_atten_pt/submit_8card.sh

@@ -0,0 +1,36 @@
+#!/bin/bash
+#SBATCH --job-name=dp_pt_atten_8card
+#SBATCH --partition=hx1hdexclu12
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=1
+#SBATCH --gres=dcu:8
+#SBATCH --cpus-per-task=16
+#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 "$SCRIPT_DIR/matchem_env.sh"
+
+# DeepMD 训练环境已由 matchem_env.sh 覆盖
+
+# 多卡训练（8卡）
+cd "$SCRIPT_DIR"
+
+# 输入 JSON 使用环境变量，运行前展开为临时文件
+INPUT_JSON="input_torch.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('$INPUT_JSON').read()))"
+
+torchrun --nproc_per_node=8 -m deepmd --pt train "$EXPANDED_JSON"

demo/water_se_e2_a_pt/input_torch.json

@@ -0,0 +1,81 @@
+{
+  "model": {
+    "type_map": [
+      "O",
+      "H"
+    ],
+    "descriptor": {
+      "type": "se_e2_a",
+      "sel": [
+        46,
+        92
+      ],
+      "rcut_smth": 0.5,
+      "rcut": 6.0,
+      "neuron": [
+        25,
+        50,
+        100
+      ],
+      "resnet_dt": false,
+      "axis_neuron": 16,
+      "type_one_side": true,
+      "seed": 1,
+      "_comment": " that's all"
+    },
+    "fitting_net": {
+      "neuron": [
+        240,
+        240,
+        240
+      ],
+      "resnet_dt": true,
+      "seed": 1,
+      "_comment": " that's all"
+    },
+    "data_stat_nbatch": 20,
+    "_comment": " that's all"
+  },
+  "learning_rate": {
+    "type": "exp",
+    "decay_steps": 5000,
+    "start_lr": 0.001,
+    "stop_lr": 3.51e-08,
+    "_comment": "that's all"
+  },
+  "loss": {
+    "type": "ener",
+    "start_pref_e": 0.02,
+    "limit_pref_e": 1,
+    "start_pref_f": 1000,
+    "limit_pref_f": 1,
+    "_comment": " that's all"
+  },
+  "training": {
+    "stat_file": "./se_e2_a.hdf5",
+    "training_data": {
+      "systems": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/dp/water/data_0",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/dp/water/data_1",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/dp/water/data_2"
+      ],
+      "batch_size": 4,
+      "_comment": "that's all"
+    },
+    "validation_data": {
+      "systems": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/dp/water/data_3"
+      ],
+      "batch_size": 4,
+      "numb_btch": 3,
+      "_comment": "that's all"
+    },
+    "numb_steps": 10000,
+    "seed": 10,
+    "disp_file": "lcurve.out",
+    "disp_freq": 1000,
+    "save_freq": 10000,
+    "_comment": "that's all"
+  },
+  "_comment": "that's all"
+}

demo/water_se_e2_a_pt/submit_1card.sh

@@ -0,0 +1,35 @@
+#!/bin/bash
+#SBATCH --job-name=dp_pt_se_e2_a
+#SBATCH --partition=hpctest02
+#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 $SCRIPT_DIR
+
+source "$SCRIPT_DIR/matchem_env.sh"
+
+# DeepMD 训练环境已由 matchem_env.sh 覆盖
+
+# 单卡训练
+cd "$SCRIPT_DIR"
+
+# 输入 JSON 使用环境变量，运行前展开为临时文件
+INPUT_JSON="input_torch.json"
+EXPANDED_JSON=".${INPUT_JSON%.json}_expanded_$$.json"
+trap 'rm -f "$EXPANDED_JSON"' EXIT
+python3 -c "import os; open('$EXPANDED_JSON','w').write(os.path.expandvars(open('$INPUT_JSON').read()))"
+
+dp --pt train "$EXPANDED_JSON"

demo/water_se_e2_a_pt/submit_4card.sh

@@ -0,0 +1,35 @@
+#!/bin/bash
+#SBATCH --job-name=dp_pt_4card
+#SBATCH --partition=hpctest02
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=1
+#SBATCH --gres=dcu:4
+#SBATCH --cpus-per-task=16
+#SBATCH --time=2:00:00
+#SBATCH --output=slurm_4card_%j.out
+#SBATCH --error=slurm_4card_%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 $SCRIPT_DIR
+
+source "$SCRIPT_DIR/matchem_env.sh"
+
+# DeepMD 训练环境已由 matchem_env.sh 覆盖
+
+# 多卡训练（4卡）
+cd "$SCRIPT_DIR"
+
+# 输入 JSON 使用环境变量，运行前展开为临时文件
+INPUT_JSON="input_torch.json"
+EXPANDED_JSON=".${INPUT_JSON%.json}_expanded_$$.json"
+trap 'rm -f "$EXPANDED_JSON"' EXIT
+python3 -c "import os; open('$EXPANDED_JSON','w').write(os.path.expandvars(open('$INPUT_JSON').read()))"
+
+torchrun --nproc_per_node=4 -m deepmd --pt train "$EXPANDED_JSON"

demo/water_se_e2_a_pt/submit_8card.sh

@@ -0,0 +1,35 @@
+#!/bin/bash
+#SBATCH --job-name=dp_pt_8card
+#SBATCH --partition=hpctest02
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=1
+#SBATCH --gres=dcu:8
+#SBATCH --cpus-per-task=16
+#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 $SCRIPT_DIR
+
+source "$SCRIPT_DIR/matchem_env.sh"
+
+# DeepMD 训练环境已由 matchem_env.sh 覆盖
+
+# 多卡训练（8卡）
+cd "$SCRIPT_DIR"
+
+# 输入 JSON 使用环境变量，运行前展开为临时文件
+INPUT_JSON="input_torch.json"
+EXPANDED_JSON=".${INPUT_JSON%.json}_expanded_$$.json"
+trap 'rm -f "$EXPANDED_JSON"' EXIT
+python3 -c "import os; open('$EXPANDED_JSON','w').write(os.path.expandvars(open('$INPUT_JSON').read()))"
+
+torchrun --nproc_per_node=8 -m deepmd --pt train "$EXPANDED_JSON"

demo/water_se_e2_a_tf/input_tf.json

@@ -0,0 +1,88 @@
+{
+  "_comment1": " model parameters",
+  "model": {
+    "type_map": [
+      "O",
+      "H"
+    ],
+    "descriptor": {
+      "type": "se_e2_a",
+      "sel": [
+        46,
+        92
+      ],
+      "rcut_smth": 0.50,
+      "rcut": 6.00,
+      "neuron": [
+        25,
+        50,
+        100
+      ],
+      "resnet_dt": false,
+      "axis_neuron": 16,
+      "type_one_side": true,
+      "precision": "float64",
+      "seed": 1,
+      "_comment2": " that's all"
+    },
+    "fitting_net": {
+      "neuron": [
+        240,
+        240,
+        240
+      ],
+      "resnet_dt": true,
+      "precision": "float64",
+      "seed": 1,
+      "_comment3": " that's all"
+    },
+    "_comment4": " that's all"
+  },
+
+  "learning_rate": {
+    "type": "exp",
+    "decay_steps": 5000,
+    "start_lr": 0.001,
+    "stop_lr": 3.51e-8,
+    "_comment5": "that's all"
+  },
+
+  "loss": {
+    "type": "ener",
+    "start_pref_e": 0.02,
+    "limit_pref_e": 1,
+    "start_pref_f": 1000,
+    "limit_pref_f": 1,
+    "start_pref_v": 0,
+    "limit_pref_v": 0,
+    "_comment6": " that's all"
+  },
+
+  "training": {
+    "training_data": {
+      "systems": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/dp/water/data_0/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/dp/water/data_1/",
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/dp/water/data_2/"
+      ],
+      "batch_size": "auto",
+      "_comment7": "that's all"
+    },
+    "validation_data": {
+      "systems": [
+        "/public/share/sugonhpcapp01/onestore/onedatasets/matchem/dp/water/data_3"
+      ],
+      "batch_size": 1,
+      "numb_btch": 3,
+      "_comment8": "that's all"
+    },
+    "numb_steps": 10000,
+    "seed": 10,
+    "disp_file": "lcurve.out",
+    "disp_freq": 1000,
+    "save_freq": 10000,
+    "_comment9": "that's all"
+  },
+
+  "_comment10": "that's all"
+}

demo/water_se_e2_a_tf/submit.sh

@@ -0,0 +1,42 @@
+#!/bin/bash
+#SBATCH --job-name=dp_tf_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 "$SCRIPT_DIR/matchem_env.sh"
+
+# DeepMD 训练环境已由 matchem_env.sh 覆盖
+
+# 限制 batch size 和线程数，规避 ROCm kernel launch 失败
+export DP_INFER_BATCH_SIZE=4096
+export DP_INTRA_OP_PARALLELISM_THREADS=1
+export DP_INTER_OP_PARALLELISM_THREADS=1
+export OMP_NUM_THREADS=1
+export TF_XLA_FLAGS=""                                                                                                                                                                                                                                     
+export TF_ENABLE_XLA=0    
+# TF 后端单卡训练
+
+# 输入 JSON 使用环境变量，运行前展开为临时文件
+INPUT_JSON="input_tf.json"
+EXPANDED_JSON=".${INPUT_JSON%.json}_expanded_$$.json"
+trap 'rm -f "$EXPANDED_JSON"' EXIT
+python3 -c "import os; open('$EXPANDED_JSON','w').write(os.path.expandvars(open('$INPUT_JSON').read()))"
+
+dp --tf train "$EXPANDED_JSON"

dp_install.sh

@@ -0,0 +1,137 @@
+#!/bin/bash
+# -----------------------------------------------------------------------------
+# DeepMD-kit DCU 一键安装脚本
+# 流程：源码拉取 → 编译安装 → 安装验证
+# 用法：bash dp_install.sh
+#       DEEPMD_SRC_DIR=/path/to/src bash dp_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 "$SCRIPT_DIR/../matchem_env.sh"
+module load sghpc-mpi-gcc/26.3
+
+# 1.5 确保 gflags/glog 运行库存在（torch cmake 与 lmp_mpi 的运行时依赖，pip 环境通常缺失）
+echo ">>> Step 1.5: 检查 gflags/glog 运行库"
+MISSING_PKGS=""
+ls "$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: $url"
+    return 1
+}
+
+# 2. 源码准备
+# 说明：
+#   - 开发/测试阶段：自动通过 HTTPS + 代理拉取源码
+#   - 生产/客户场景：建议提前上传源码到集群，通过 DEEPMD_SRC_DIR 指定
+DEEPMD_SRC="${DEEPMD_SRC_DIR:-${SCRIPT_DIR}/deepmd-kit}"
+if [ ! -d "$DEEPMD_SRC/.git" ] && [ ! -f "$DEEPMD_SRC/setup.py" ]; then
+    echo ">>> Step 2: 拉取 DeepMD-kit 源码"
+    # 当前集群需通过 HTTP 代理访问外网，配置 git 代理
+    git config --global http.proxy "http://jsyadmin:1cdf8f60@10.13.17.166:3128"
+    git clone --depth 1 "https://oauth2:${GITEE_TOKEN}@gitee.com/wang-rui-sugon/deepmd-kit_dcu.git" "$DEEPMD_SRC"
+else
+    echo ">>> Step 2: 源码已存在，跳过拉取"
+fi
+
+# 3. 预先锁定 numpy 版本，避免 deepmd-kit 安装过程中短暂升级到不兼容版本
+echo ">>> Step 3: 预先锁定 numpy 版本"
+pip install numpy==1.26.3 --no-deps -i https://pypi.tuna.tsinghua.edu.cn/simple --trusted-host pypi.tuna.tsinghua.edu.cn
+
+# 4. 修复 Torch cmake 硬编码 DTK 路径
+echo ">>> Step 4: 修复 Torch cmake 硬编码 DTK 路径"
+TORCH_PATH=$(python -c "import importlib.util, os; spec = importlib.util.find_spec('torch'); print(os.path.dirname(spec.origin) if spec and spec.origin else '')")
+CAFFE2_CMAKE="${TORCH_PATH}/share/cmake/Caffe2/Caffe2Targets.cmake"
+DTK_REAL_PATH="/public/software/sghpc_sdk.bak/Linux_x86_64/26.3/dtk/dtk-25.04.4"
+if [ -f "$CAFFE2_CMAKE" ] && grep -q '/opt/dtk' "$CAFFE2_CMAKE"; then
+    echo ">>> 替换 Caffe2Targets.cmake 中的 /opt/dtk 为实际路径"
+    sed -i "s|/opt/dtk|${DTK_REAL_PATH}|g" "$CAFFE2_CMAKE"
+fi
+
+# 5. 编译安装（PyTorch + TensorFlow 双后端）
+echo ">>> Step 5: 编译安装 Python 包（PyTorch + TensorFlow 双后端）"
+cd "$DEEPMD_SRC"
+DP_VARIANT=rocm \
+ROCM_ROOT="$ROCM_PATH" \
+DP_ENABLE_TENSORFLOW=1 \
+DP_ENABLE_PYTORCH=1 \
+PYTORCH_ROOT="${TORCH_PATH}" \
+    pip install . "numpy==1.26.3" -i https://pypi.tuna.tsinghua.edu.cn/simple --trusted-host pypi.tuna.tsinghua.edu.cn
+
+# 6. 验证
+echo ">>> Step 6: 验证安装"
+dp -h | head -n 5
+
+echo "========================================"
+echo " DeepMD-kit Python 包安装完成"
+echo "========================================"
+
+# 7. C++ 接口安装（含 LAMMPS 插件）
+# 说明：默认从预编译包下载解压，快速部署；如需自行源码编译，设置 COMPILE_DP_CPP=1。
+DP_CPP_URL="https://download.sourcefind.cn:65024/file/9/onesicence/dtk-25.04.2/deep_lammps/dp_cpp_dcu.tar.gz"
+
+if [ "${COMPILE_DP_CPP:-0}" = "1" ]; then
+    echo ">>> Step 7: 源码编译 C++ 接口（含 LAMMPS 插件）"
+
+    # 7.1 Patch Gelu op：TensorFlow 2.18+ 已内置 Gelu，与 deepmd-kit 自定义 op 冲突，
+    #     需在编译前注释掉 source/op/tf/gelu_multi_device.cc 中的 REGISTER_OP("Gelu")
+    #     和 REGISTER_OP("GeluGrad") 及其属性链。（GeluGradGrad / GeluCustom 系列不受影响）
+    GELU_FILE="$DEEPMD_SRC/source/op/tf/gelu_multi_device.cc"
+    if grep -q '^REGISTER_OP("Gelu")' "$GELU_FILE"; then
+        echo ">>> Step 7.1: Patch Gelu op 注册，避免 TF 2.18+ 冲突"
+        sed -i '/^REGISTER_OP("Gelu")$/,/^);$/{ /^$/!s/^/\/\/ /; }' "$GELU_FILE"
+        sed -i '/^REGISTER_OP("GeluGrad")$/,/^);$/{ /^$/!s/^/\/\/ /; }' "$GELU_FILE"
+    fi
+
+    cd "$DEEPMD_SRC/source"
+    mkdir -p build && cd build
+
+    cmake -DENABLE_TENSORFLOW=ON \
+          -DENABLE_PYTORCH=ON \
+          -DUSE_ROCM_TOOLKIT=ON \
+          -DTENSORFLOW_ROOT="${CONDA_PREFIX}/lib/python3.11/site-packages/tensorflow" \
+          -DTensorFlow_INCLUDE_DIRS="${CONDA_PREFIX}/lib/python3.11/site-packages/tensorflow/include" \
+          -DTorch_DIR="${CONDA_PREFIX}/lib/python3.11/site-packages/torch/share/cmake/Torch" \
+          -DHIP_ROOT_DIR="${ROCM_PATH}/hip" \
+          -DCMAKE_PREFIX_PATH="${CONDA_PREFIX};${ROCM_PATH}/lib/cmake" \
+          -DLAMMPS_SOURCE_ROOT="${LAMMPS_SRC_DIR}" \
+          -DCMAKE_INSTALL_PREFIX="${DP_CPP_DIR}" \
+          ..
+
+    make -j$(nproc)
+    make install
+
+    # 后处理：创建 dpplugin.so 符号链接
+    cd "${DP_CPP_DIR}/lib"
+    if [ -f "deepmd_lmp/dpplugin.so" ] && [ ! -e "dpplugin.so" ]; then
+        ln -s deepmd_lmp/dpplugin.so ./
+    fi
+else
+    echo ">>> Step 7: 下载预编译 C++ 接口包"
+    mkdir -p "${DP_CPP_DIR}"
+    cd "${DP_CPP_DIR}"
+    download_file "${DP_CPP_URL}" dp_cpp_dcu.tar.gz || exit 1
+    tar -xzf dp_cpp_dcu.tar.gz --strip-components=1
+    rm -f dp_cpp_dcu.tar.gz
+    echo ">>> Step 7: C++ 接口安装完成（${DP_CPP_DIR}）"
+fi

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}"