RuntimeError: unsupported scalarType

[xueking1]

Dear Hailanyi:
When I was training with Virconv-L, I found an incredible problem:
[Exception|implicit_gemm_backward]feat=torch.Size([0]),w=torch.Size([32, 3, 3, 32]),pair=torch.Size([9, 166197]),issubm=True,do=torch.Size([166197, 32]) epochs: 24%|██▍ | 12/50 [6:29:43<20:34:06, 1948.59s/it, loss=2.18, lr=0.00698] Traceback (most recent call last): File "/media/xd/hpc/xyy/VirConv-master/tools/train_utils/train_utils.py", line 95, in train_model accumulated_iter = train_one_epoch( File "/media/xd/hpc/xyy/VirConv-master/tools/train_utils/train_utils.py", line 47, in train_one_epoch loss.backward() SPCONV_DEBUG_SAVE_PATH not found, you can specify SPCONV_DEBUG_SAVE_PATH as debug data save path to save debug data which can be attached in a issue. File "/home/xd/anaconda3/envs/vircompletion/lib/python3.8/site-packages/torch/_tensor.py", line 307, in backward torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs) File "/home/xd/anaconda3/envs/vircompletion/lib/python3.8/site-packages/torch/autograd/__init__.py", line 154, in backward Variable._execution_engine.run_backward( File "/home/xd/anaconda3/envs/vircompletion/lib/python3.8/site-packages/torch/autograd/function.py", line 199, in apply return user_fn(self, *args) File "/home/xd/anaconda3/envs/vircompletion/lib/python3.8/site-packages/torch/autograd/function.py", line 340, in wrapper outputs = fn(ctx, *args) File "/home/xd/anaconda3/envs/vircompletion/lib/python3.8/site-packages/torch/cuda/amp/autocast_mode.py", line 111, in decorate_bwd return bwd(*args, **kwargs) File "/home/xd/anaconda3/envs/vircompletion/lib/python3.8/site-packages/spconv/pytorch/functional.py", line 258, in backward raise e File "/home/xd/anaconda3/envs/vircompletion/lib/python3.8/site-packages/spconv/pytorch/functional.py", line 234, in backward input_bp, filters_bp = ops.implicit_gemm_backward( File "/home/xd/anaconda3/envs/vircompletion/lib/python3.8/site-packages/spconv/pytorch/ops.py", line 1188, in implicit_gemm_backward if features.dtype == torch.int8 or features.dtype == torch.qint8: RuntimeError: unsupported scalarType python-BaseException

But when I trained Virconv-S, I found no such problems. Have you ever encountered such a problem before? Thank you!