hi ,tanaks for your excelent work!
when I run this code: ” out_feat = assign_score_cuda(scores, kernel_feat, half_kernel_feat, grouped_idx, aggregate='sum') “ in paconv.py,
I got the output of parameter 'out_feat' as shown in the figure below, the data of parameter 'out_feat' cannot be displayed.
I have no idea what went wrong, could you give some explanation ?
Thanks a lot!
out_feat
<torch.Tensor object at 0x7fb5941218b0>
special variables
function variables
H:
'Traceback (most recent call last):\n File "../.vscode-server/extensions/ms-python.debugpy-2024.1.10371006/bundled/libs/debugpy/_vendored/pydevd/_pydevd_bundle/pydevd_resolver.py", line 189, in _get_py_dictionary\n attr = getattr(var, name)\nRuntimeError: tensor.H is only supported on matrices (2-D tensors). Got 3-D tensor. For batches of matrices, consider using tensor.mH\n'
T:
<torch.Tensor object at 0x7fb5865f61d0>
data:
<torch.Tensor object at 0x7fb6de4ecf90> ( look at this =) )
device:
device(type='cuda', index=0)
dtype:
torch.float32
grad:
None
grad_fn:
<torch.autograd.function.AssignScoreWithKBackward object at 0x7fb59402b400>
imag:
'Traceback (most recent call last):\n File "../.vscode-server/extensions/ms-python.debugpy-2024.1.10371006/bundled/libs/debugpy/_vendored/pydevd/_pydevd_bundle/pydevd_resolver.py", line 189, in _get_py_dictionary\n attr = getattr(var, name)\nRuntimeError: imag is not implemented for tensors with non-complex dtypes.\n'
is_cuda:
True
is_ipu:
False
is_leaf:
False
is_meta:
False
is_mkldnn:
False
is_mps:
False
is_nested:
False
is_ort:
False
is_quantized:
False
is_sparse:
False
is_sparse_csr:
False
is_vulkan:
False
is_xpu:
False
layout:
torch.strided
mH:
<torch.Tensor object at 0x7fb6de4f0a40>
mT:
<torch.Tensor object at 0x7fb6de4f0a90>
name:
None
names:
(None, None, None)
ndim:
3
output_nr:
0
real:
<torch.Tensor object at 0x7fb5941218b0>
requires_grad:
True
retains_grad:
False
shape:
torch.Size([2, 128, 512])
_addmm_activation:
<built-in method _addmm_activation of Tensor object at 0x7fb5941218b0>
_autocast_to_full_precision:
<built-in method _autocast_to_full_precision of Tensor object at 0x7fb5941218b0>
_autocast_to_reduced_precision:
<built-in method _autocast_to_reduced_precision of Tensor object at 0x7fb5941218b0>
_backward_hooks:
None
_base:
None
_cdata:
94437083738576
coalesced:
<built-in method coalesced of Tensor object at 0x7fb5941218b0>
_conj:
<built-in method _conj of Tensor object at 0x7fb5941218b0>
_conj_physical:
<built-in method _conj_physical of Tensor object at 0x7fb5941218b0>
_dimI:
<built-in method _dimI of Tensor object at 0x7fb5941218b0>
hi ,tanaks for your excelent work! when I run this code: ” out_feat = assign_score_cuda(scores, kernel_feat, half_kernel_feat, grouped_idx, aggregate='sum') “ in paconv.py,
I got the output of parameter 'out_feat' as shown in the figure below, the data of parameter 'out_feat' cannot be displayed.
I have no idea what went wrong, could you give some explanation ? Thanks a lot!
out_feat <torch.Tensor object at 0x7fb5941218b0> special variables function variables H: 'Traceback (most recent call last):\n File "../.vscode-server/extensions/ms-python.debugpy-2024.1.10371006/bundled/libs/debugpy/_vendored/pydevd/_pydevd_bundle/pydevd_resolver.py", line 189, in _get_py_dictionary\n attr = getattr(var, name)\nRuntimeError: tensor.H is only supported on matrices (2-D tensors). Got 3-D tensor. For batches of matrices, consider using tensor.mH\n' T: <torch.Tensor object at 0x7fb5865f61d0> data: <torch.Tensor object at 0x7fb6de4ecf90> ( look at this =) ) device: device(type='cuda', index=0) dtype: torch.float32 grad: None grad_fn: <torch.autograd.function.AssignScoreWithKBackward object at 0x7fb59402b400> imag: 'Traceback (most recent call last):\n File "../.vscode-server/extensions/ms-python.debugpy-2024.1.10371006/bundled/libs/debugpy/_vendored/pydevd/_pydevd_bundle/pydevd_resolver.py", line 189, in _get_py_dictionary\n attr = getattr(var, name)\nRuntimeError: imag is not implemented for tensors with non-complex dtypes.\n' is_cuda: True is_ipu: False is_leaf: False is_meta: False is_mkldnn: False is_mps: False is_nested: False is_ort: False is_quantized: False is_sparse: False is_sparse_csr: False is_vulkan: False is_xpu: False layout: torch.strided mH: <torch.Tensor object at 0x7fb6de4f0a40> mT: <torch.Tensor object at 0x7fb6de4f0a90> name: None names: (None, None, None) ndim: 3 output_nr: 0 real: <torch.Tensor object at 0x7fb5941218b0> requires_grad: True retains_grad: False shape: torch.Size([2, 128, 512]) _addmm_activation: <built-in method _addmm_activation of Tensor object at 0x7fb5941218b0> _autocast_to_full_precision: <built-in method _autocast_to_full_precision of Tensor object at 0x7fb5941218b0> _autocast_to_reduced_precision: <built-in method _autocast_to_reduced_precision of Tensor object at 0x7fb5941218b0> _backward_hooks: None _base: None _cdata: 94437083738576 coalesced: <built-in method coalesced of Tensor object at 0x7fb5941218b0> _conj: <built-in method _conj of Tensor object at 0x7fb5941218b0> _conj_physical: <built-in method _conj_physical of Tensor object at 0x7fb5941218b0> _dimI: <built-in method _dimI of Tensor object at 0x7fb5941218b0>