Floating point exception

[rhapsody96]

I reproduced your work in the following environment:
Nvidia H20
python==3.11.14
torch=2.3.1
torchvision==0.18.1

When I run: python main.py --image_folder ./office_loop --max_loops 1 --vis_map but have problems with it, do you have any ideas?

Using device: cuda
Starting viser server on port 8080
/VGGT-SLAM/salad/salad/models_salad/backbones/dinov2.py:34: FutureWarning: You are using torch.load with weights_only=False (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details).In a future release, the default value for weights_only will be flipped to True. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via torch.serialization.add_safe_globals. We recommend you start setting weights_only=True for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
self.model.load_state_dict(torch.load('/VGGT-SLAM/dinov2_vitb14_pretrain.pth'))
Loaded model from /VGGT-SLAM/dino_salad.ckpt Successfully!
Starting viser server...
Initializing and loading VGGT model...
/VGGT-SLAM/main.py:62: FutureWarning: You are using torch.load with weights_only=False (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details).In a future release, the default value for weights_only will be flipped to True. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via torch.serialization.add_safe_globals. We recommend you start setting weights_only=True for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
model.load_state_dict(torch.load(_URL))
Loading images from office_loop...
Found 473 images
12%|███████████████████████████ | 55/473 [00:00<00:06, 60.32
it/s]['office_loop/frame_0001.jpg', 'office_loop/frame_0010.jpg', 'office_loop/frame_0014.jpg', 'office_loop/frame_0018.jpg', 'office_loop/frame_0021.jpg', 'office_loop/frame_0024.jpg', 'office_loop/frame_0026.jpg', 'office_loop/frame_0029.jpg', 'office_loop/frame_0032.jpg', 'office_loop/frame_0035.jpg', 'office_loop/frame_0040.jpg', 'office_loop/frame_0044.jpg', 'office_loop/frame_0048.jpg', 'office_loop/frame_0052.jpg', 'office_loop/frame_0055.jpg', 'office_loop/frame_0058.jpg', 'office_loop/frame_0060.jpg']
Preprocessed images shape: torch.Size([17, 3, 294, 518])
/VGGT-SLAM/vggt_slam/solver.py:433: FutureWarning: torch.cuda.amp.autocast(args...) is deprecated. Please use torch.amp.autocast('cuda', args...) instead.
with torch.cuda.amp.autocast(dtype=dtype):
/VGGT-SLAM/vggt/vggt/models/vggt.py:65: FutureWarning: torch.cuda.amp.autocast(args...) is deprecated. Please use torch.amp.autocast('cuda', args...) instead.
with torch.cuda.amp.autocast(enabled=False):
Fatal Python error: Floating point exception

Current thread 0x00007fa87c5e4740 (most recent call first):
File "/anaconda/envs/vggt-slam/lib/python3.11/site-packages/torch/nn/modules/linear.py", line 117 in forward
File "/VGGT-SLAM/vggt/vggt/layers/mlp.py", line 35 in forward
x = self.fc1(x)