Problems with various versions of gpu4pyscf

Hello,

I am trying to use this library to compute nonadiabatic couplings between TDDFT excited states. Up to version 1.4.3, whenever I try to do a TD calculation, I get

File ~/anaconda3/envs/working/lib/python3.13/site-packages/gpu4pyscf/lib/cutensor.py:91, in contraction(pattern, a, b, alpha, beta, out, op_a, op_b, op_c, algo, jit_mode, compute_desc, ws_pref)
     88     out = cupy.empty([shape[k] for k in str_c], order='C', dtype=dtype)
     89 c = out
---> 91 desc_a = cutensor.create_tensor_descriptor(a)
     92 desc_b = cutensor.create_tensor_descriptor(b)
     93 desc_c = cutensor.create_tensor_descriptor(c)

File cupyx/cutensor.pyx:235, in cupyx.cutensor.create_tensor_descriptor()

File cupyx/cutensor.pyx:244, in cupyx.cutensor.create_tensor_descriptor()

File cupyx/cutensor.pyx:207, in cupyx.cutensor._get_handle()

File cupyx/cutensor.pyx:55, in cupyx.cutensor.Handle.__init__()

File cupy_backends/cuda/libs/cutensor.pyx:200, in cupy_backends.cuda.libs.cutensor.create()

File cupy_backends/cuda/libs/cutensor.pyx:189, in cupy_backends.cuda.libs.cutensor.check_status()

CuTensorError: CUTENSOR_STATUS_NOT_SUPPORTED

On the other side, if I install the latest 1.5.x versions, I obtain the same message as reported here.

I'd like to solve at least one issue so I can use one of the two versions to carry out the calculations I need, any help would be appreciated.

Thanks in advance

Start with the traceback at gpu4pyscf/lib/cutensor.py:91 and compare the environments for versions 1.4.3 and 1.5.x. Review issue 595 and reproduce the TD calculation to determine whether either version can run with a supported dependency combination. Done means one affected version has a reproducible resolution or the incompatibility is documented clearly.