Segmentation fault (core dumped)

[luuude]

Hi I am getting segmentation fault. Any idea why?

python3 gaussian_point_train.py --train_config config/tat_truck_every_8_test.yaml

[Taichi] version 1.6.0, llvm 15.0.4, commit f1c6fbbd, linux, python 3.10.12
[Taichi] Starting on arch=cuda
0%| | 0/30001 [00:00<?, ?it/s]/home/ai/.local/lib/python3.10/site-packages/torchvision/transforms/functional.py:1603: UserWarning: The default value of the antialias parameter of all the resizing transforms (Resize(), RandomResizedCrop(), etc.) will change from None to True in v0.17, in order to be consistent across the PIL and Tensor backends. To suppress this warning, directly pass antialias=True (recommended, future default), antialias=None (current default, which means False for Tensors and True for PIL), or antialias=False (only works on Tensors - PIL will still use antialiasing). This also applies if you are using the inference transforms from the models weights: update the call to weights.transforms(antialias=True).
warnings.warn(
/home/ai/.local/lib/python3.10/site-packages/taichi/lang/expr.py:101: DeprecationWarning: In future, it will be an error for 'np.bool_' scalars to be interpreted as an index
return Expr(_ti_core.make_const_expr_int(constant_dtype, val))
0%| | 11/30001 [00:04<1:50:52, 4.51it/s]Segmentation fault (core dumped)

[jb-ye]

I had exactly segfault issue with the latest commit and had to revert this commit to proceed.

[wanmeihuali]

already roll back to previous commit, the problem shall be fix now