When using the DP deep learning force field model graph.pb for energy and force calculations on structures, energy and force inference fails, with NAN as the output

[Zhou-tao-USTC]

1：It never works when running on a single core.
2：When running on multiple cores, it sometimes successfully infers energy and force, while other times it fails.
3：Even when setting DP_INFER_BATCH_SIZE=1, it still doesn't work during single-core operation.
Uploading test.zip…

[wanghan-iapcm]

from your code

model_path = "../graph.pb"
try:
    model = DeepPot(model_path)
    supported_types = model.get_type_map()
    print(f"模型支持的原子类型: {supported_types}")
except Exception as e:
    print(f"模型加载失败: {e}")
    exit()

# 3. 设置计算器
try:
    atoms.calc = DP(model=model_path, type_map=supported_types)
    print("DeepMD计算器创建成功")
except Exception as e:
    print(f"计算器创建失败: {e}")
    exit()

it is not correct to set type_map=supported_types, as the supproted_types may be printed in any order.

please try not explicitly setting type_map

or if you really want to set a type_map, you should explicitly know how map ase atom names to the corresponding DP atom types.

[Zhou-tao-USTC]

But it seems that the problem still hasn't been resolved.

There's one point I don't understand: When running with multiple cores, sometimes the energy and forces can be output normally (test-2-parallel-2), but when running on a single core, no matter how I try, the results are always NAN.

I've modified the test script to not use the type_map approach as per your suggestion.

The final output log is as follows:
loading conda env : /work/home/zhoutao_2024/apprepo/deepmd_cpu/v3.0.0b0-none/app/deepmd-kit-3.0.0b0-cpu-Linux-x86_64 .....
2025-09-01 10:02:59.182193: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: SSE4.1 SSE4.2 AVX AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
WARNING:tensorflow:From /work/home/zhoutao_2024/apprepo/deepmd_cpu/v3.0.0b0-none/app/deepmd-kit-3.0.0b0-cpu-Linux-x86_64/lib/python3.11/site-packages/tensorflow/python/compat/v2_compat.py:108: disable_resource_variables (from tensorflow.python.ops.variable_scope) is deprecated and will be removed in a future version.
Instructions for updating:
non-resource variables are not supported in the long term
2025-09-01 10:06:03.653599: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:388] MLIR V1 optimization pass is not enabled
WARNING:tensorflow:From /work/home/zhoutao_2024/apprepo/deepmd_cpu/v3.0.0b0-none/app/deepmd-kit-3.0.0b0-cpu-Linux-x86_64/lib/python3.11/site-packages/deepmd/tf/utils/batch_size.py:28: is_gpu_available (from tensorflow.python.framework.test_util) is deprecated and will be removed in a future version.
Instructions for updating:
Use tf.config.list_physical_devices('GPU') instead.
You can use the environment variable DP_INFER_BATCH_SIZE tocontrol the inference batch size (nframes * natoms). The default value is 1024.
You can use the environment variable DP_INFER_BATCH_SIZE tocontrol the inference batch size (nframes * natoms). The default value is 1024.

===== 尝试1: 原始结构 =====
结构原子数: 162
模型支持的原子类型: ['O', 'Al', 'H']
DeepMD计算器创建成功
能量: nan eV
最大力: nan eV/Å
[警告] 检测到力计算结果包含NaN值!

Uploading CONTCAR.vasp…

[njzjz]

Cannot reproduce:

python test.py     

===== 尝试1: 原始结构 =====
结构原子数: 162
To get the best performance, it is recommended to adjust the number of threads by setting the environment variables OMP_NUM_THREADS, DP_INTRA_OP_PARALLELISM_THREADS, and DP_INTER_OP_PARALLELISM_THREADS. See https://deepmd.rtfd.io/parallelism/ for more information.
2025-09-01 11:36:24.544541: I metal_plugin/src/device/metal_device.cc:1154] Metal device set to: Apple M4 Pro
2025-09-01 11:36:24.544576: I metal_plugin/src/device/metal_device.cc:296] systemMemory: 48.00 GB
2025-09-01 11:36:24.544587: I metal_plugin/src/device/metal_device.cc:313] maxCacheSize: 18.72 GB
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1756697784.544607 34830653 pluggable_device_factory.cc:305] Could not identify NUMA node of platform GPU ID 0, defaulting to 0. Your kernel may not have been built with NUMA support.
I0000 00:00:1756697784.544630 34830653 pluggable_device_factory.cc:271] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 0 MB memory) -> physical PluggableDevice (device: 0, name: METAL, pci bus id: <undefined>)
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1756697784.552729 34830653 mlir_graph_optimization_pass.cc:425] MLIR V1 optimization pass is not enabled
2025-09-01 11:36:24.560405: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:117] Plugin optimizer for device_type GPU is enabled.
模型支持的原子类型: ['O', 'Al', 'H']
I0000 00:00:1756697784.640179 34830653 pluggable_device_factory.cc:305] Could not identify NUMA node of platform GPU ID 0, defaulting to 0. Your kernel may not have been built with NUMA support.
I0000 00:00:1756697784.640202 34830653 pluggable_device_factory.cc:271] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 0 MB memory) -> physical PluggableDevice (device: 0, name: METAL, pci bus id: <undefined>)
DeepMD计算器创建成功
能量: -45585.186914 eV
最大力: 0.518222 eV/Å