[BUG] <can not use fp16>

[momoyo126]

Is there an existing issue for this?

• I have searched the existing issues

Current Behavior

Although conv_mode has been set to 0, but the model still uses fp32. I try to ues amp.autocast but model still uses fp16. But if using torchsparse2.0.0, there is no problem.

Expected Behavior

No response

Environment

- GCC: 11.4
- NVCC: 11.8
- PyTorch: 2.1
- PyTorch CUDA: 11.8
- TorchSparse: 2.1

Anything else?

No response

[francotheengineer]

Please provide minimal code example. Creating issues with bare minimum effort means we cannot help you!