[BUG] `dp eval-desc` ValueError when deal with `deepmd/npy/mixed` format contain multiple frames

[QuantumMisaka]

Bug summary

When using dp --pt eval-desc -m DPA-3.1-3M.pt -s ./18 --head OC20Mfor adeepmd/npy/mixed` MultiSystems having multiple frames, the eval-desc process will ValueError

dp --pt eval-desc -m ../DPA-3.1-3M.pt -s 18 --head OC20M
[2026-01-26 22:52:25,695] DEEPMD WARNING 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.
[2026-01-26 22:52:30,346] DEEPMD WARNING You can use the environment variable DP_INFER_BATCH_SIZE tocontrol the inference batch size (nframes * natoms). The default value is 1024.
[2026-01-26 22:52:31,951] DEEPMD WARNING You can use the environment variable DP_INFER_BATCH_SIZE tocontrol the inference batch size (nframes * natoms). The default value is 1024.
[2026-01-26 22:52:31,952] DEEPMD INFO    # -------output of dp eval_desc------- 
[2026-01-26 22:52:31,952] DEEPMD INFO    # processing system : 18
[2026-01-26 22:52:32,149] DEEPMD INFO    # evaluating descriptors for 83 frames
Traceback (most recent call last):
  File "/home/pku-jianghong/liuzhaoqing/.conda/envs/dpeva-dev/bin/dp", line 9, in <module>
    sys.exit(main())
             ^^^^^^
  File "/home/pku-jianghong/liuzhaoqing/.conda/envs/dpeva-dev/lib/python3.12/site-packages/deepmd/main.py", line 1020, in main
    deepmd_main(args)
  File "/home/pku-jianghong/liuzhaoqing/.conda/envs/dpeva-dev/lib/python3.12/site-packages/deepmd/entrypoints/main.py", line 78, in main
    eval_desc(**dict_args)
  File "/home/pku-jianghong/liuzhaoqing/.conda/envs/dpeva-dev/lib/python3.12/site-packages/deepmd/entrypoints/eval_desc.py", line 126, in eval_desc
    descriptors = dp.eval_descriptor(
                  ^^^^^^^^^^^^^^^^^^^
  File "/home/pku-jianghong/liuzhaoqing/.conda/envs/dpeva-dev/lib/python3.12/site-packages/deepmd/infer/deep_eval.py", line 528, in eval_descriptor
    descriptor = self.deep_eval.eval_descriptor(
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/pku-jianghong/liuzhaoqing/.conda/envs/dpeva-dev/lib/python3.12/site-packages/deepmd/pt/infer/deep_eval.py", line 782, in eval_descriptor
    self.eval(
  File "/home/pku-jianghong/liuzhaoqing/.conda/envs/dpeva-dev/lib/python3.12/site-packages/deepmd/pt/infer/deep_eval.py", line 362, in eval
    out = self._eval_func(self._eval_model, numb_test, natoms)(
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/pku-jianghong/liuzhaoqing/.conda/envs/dpeva-dev/lib/python3.12/site-packages/deepmd/pt/infer/deep_eval.py", line 442, in eval_func
    return self.auto_batch_size.execute_all(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/pku-jianghong/liuzhaoqing/.conda/envs/dpeva-dev/lib/python3.12/site-packages/deepmd/utils/batch_size.py", line 208, in execute_all
    n_batch, result = self.execute(execute_with_batch_size, index, natoms)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/pku-jianghong/liuzhaoqing/.conda/envs/dpeva-dev/lib/python3.12/site-packages/deepmd/utils/batch_size.py", line 112, in execute
    raise e
  File "/home/pku-jianghong/liuzhaoqing/.conda/envs/dpeva-dev/lib/python3.12/site-packages/deepmd/utils/batch_size.py", line 109, in execute
    n_batch, result = callable(max(batch_nframes, 1), start_index)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/pku-jianghong/liuzhaoqing/.conda/envs/dpeva-dev/lib/python3.12/site-packages/deepmd/utils/batch_size.py", line 179, in execute_with_batch_size
    return (end_index - start_index), callable(
                                      ^^^^^^^^^
  File "/home/pku-jianghong/liuzhaoqing/.conda/envs/dpeva-dev/lib/python3.12/site-packages/deepmd/pt/infer/deep_eval.py", line 487, in _eval_model
    coords.reshape([nframes, natoms, 3]).astype(prec),
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ValueError: cannot reshape array of size 54 into shape (1,1494,3)

DeePMD-kit Version

3.1.2

Backend and its version

Pytorch

How did you download the software?

Built from source

Input Files, Running Commands, Error Log, etc.

18.tar.gz

Steps to Reproduce

using dp --pt eval-desc -m DPA-3.1-3M.pt -s ./18 --head OC20M`

Further Information, Files, and Links

No response

[dosubot]

This ValueError is almost always caused by a mismatch between the expected and actual shape of your input arrays—typically, inconsistent frame or atom counts across your .npy files, or a label mismatch (like storing global tensors in atomic tensor files or vice versa) source. For deepmd/npy/mixed multi-frame systems, all frames must have the same number of atoms and consistent atom types source.

Here’s a quick checklist:

• Make sure every .npy file in your system directory has the same number of frames.
• Confirm that all frames have the same number of atoms and atom types.
• Double-check that you haven’t mixed up atomic and global tensor files.

Workarounds that have helped others:

• Try running with a smaller number of frames using the -n parameter (e.g., -n 10). This can sometimes bypass the error if it’s batch-related source.
• Set the environment variable DP_INFER_BATCH_SIZE to a lower value to control the inference batch size.
• If you have control over the code, commenting out torch.jit.script in the PyTorch model evaluation code has made similar errors disappear for some users—this points to a possible PyTorch JIT bug source.
• For very large systems, consider using LAMMPS’s rerun command with compute deeptensor/atom as a workaround source.

If you’re still stuck, try processing each frame or system individually and concatenating the results, as this can sidestep batch processing issues.

Let me know if you want more details on any of these steps!

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}