I tried to write a neural network with pytorch 1.8 cudatool 11.1 on RTX4090. But I met the following issues.

MAGMA error: function-specific error, see documentation (10) in magma_ssyevd_gpu at /opt/conda/conda-bld/magma-cuda111_1605822518874/work/src/ssyevd_gpu.cpp:226
MAGMA error: function-specific error, see documentation (10) in magma_ssyevd_gpu at /opt/conda/conda-bld/magma-cuda111_1605822518874/work/src/ssyevd_gpu.cpp:226
MAGMA error: function-specific error, see documentation (10) in magma_spotrf_LL_expert_gpu at /opt/conda/conda-bld/magma-cuda111_1605822518874/work/src/spotrf_gpu.cpp:137
MAGMA error: function-specific error, see documentation (10) in magma_spotrf_LL_expert_gpu at /opt/conda/conda-bld/magma-cuda111_1605822518874/work/src/spotrf_gpu.cpp:138
Traceback (most recent call last):
  File "baselines_locfin_nontrainable.py", line 119, in <module>
    evaluate_nontrainable_policy_locfin(
  File "baselines_locfin_nontrainable.py", line 82, in evaluate_nontrainable_policy_locfin
    [-pce_loss_lower.loss() for _ in range(n_rollout)]
  File "baselines_locfin_nontrainable.py", line 82, in <listcomp>
    [-pce_loss_lower.loss() for _ in range(n_rollout)]
  File "/home/grads/g/g.lin/Documents/idad-main/estimators/mi.py", line 134, in loss
    loss_to_constant = torch_item(self.differentiable_loss(*args, **kwargs))
  File "/home/grads/g/g.lin/Documents/idad-main/estimators/mi.py", line 111, in differentiable_loss
    primary_trace = self.get_primary_rollout(args, kwargs)
  File "/home/grads/g/g.lin/Documents/idad-main/estimators/mi.py", line 53, in get_primary_rollout
    trace.compute_log_prob()
  File "/home/grads/g/g.lin/miniconda3/envs/idad_code/lib/python3.8/site-packages/pyro/poutine/trace_struct.py", line 216, in compute_log_prob
    log_p = site["fn"].log_prob(site["value"], *site["args"], **site["kwargs"])
  File "/home/grads/g/g.lin/miniconda3/envs/idad_code/lib/python3.8/site-packages/torch/distributions/independent.py", line 91, in log_prob
    log_prob = self.base_dist.log_prob(value)
  File "/home/grads/g/g.lin/miniconda3/envs/idad_code/lib/python3.8/site-packages/torch/distributions/multivariate_normal.py", line 210, in log_prob
    M = _batch_mahalanobis(self._unbroadcasted_scale_tril, diff)
  File "/home/grads/g/g.lin/miniconda3/envs/idad_code/lib/python3.8/site-packages/torch/distributions/multivariate_normal.py", line 57, in _batch_mahalanobis
    M_swap = torch.triangular_solve(flat_x_swap, flat_L, upper=False)[0].pow(2).sum(-2)  # shape = b x c
RuntimeError: CUDA error: operation not supported when calling `cusparseCreate(handle)`

I tried to update my pytorch and cudatool version.but it will lead different issues.