mit-han-lab / torchsparse

[MICRO'23, MLSys'22] TorchSparse: Efficient Training and Inference Framework for Sparse Convolution on GPUs.

Home Page:https://torchsparse.mit.edu

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Not support DDP training

pycoco opened this issue · comments

after torch.nn.SyncBatchNorm.convert_sync_batchnorm(model), get error: AttributeError: 'SparseTensor' object has no attribute 'dim'

Hi! You may need to change the function calling when refer to the dimension of the SparseTensor. For example, you can use SparseTensor.C.shape to get the coordinates' shape and SparseTensor.F.shape to get the feature shape. This is because SparseTensor has more complicated wrapper to record the information of sparsity, compared to the dense tensor.

Close as completed due to inactivity. If you have further questions, please feel free to reopen it.