Torch Scatter with Pytorch 2.4.1 and Cu118

[tathya-ow]

Hey Team,

I have been trying to install the MeshCoder for inference and for some reason torch scatter does not have the supported version on torch 2.4.1 and Cuda 11.8. I am wondering how did you guys manage to make it work as I am getting errror by using the given pip install torch-scatter -f https://data.pyg.org/whl/torch-2.4.1%2Bcu118.html, as you look closely while installing, the screenshot shows that torch-scatter uses torch 2.4.0 as the cached version.

Could you help me figure my way out?

The error is

Traceback (most recent call last):
 File "/mnt/o/share/Tathya/MeshCoder/recipes/inference/local_inference/custom_inference/inference_shape2code_point_input.py", line 20, in <module>
 from llama_recipes.custom_models.modules.shape_tokenizer import ShapeTokenizer
 File "/mnt/o/share/Tathya/MeshCoder/src/llama_recipes/custom_models/modules/shape_tokenizer.py", line 1, in <module>
 from llama_recipes.custom_models.modules.pcd_proj.point_projection import PointTriplaneProjector
 File "/mnt/o/share/Tathya/MeshCoder/src/llama_recipes/custom_models/modules/pcd_proj/point_projection.py", line 4, in <module>
 from torch_scatter import scatter, scatter_mean, scatter_max
 File "/home/maint/miniforge3/envs/meshcoder/lib/python3.10/site-packages/torch_scatter/__init__.py", line 16, in <module>
 torch.ops.load_library(spec.origin)
 File "/home/maint/miniforge3/envs/meshcoder/lib/python3.10/site-packages/torch/_ops.py", line 1295, in load_library
 ctypes.CDLL(path)
 File "/home/maint/miniforge3/envs/meshcoder/lib/python3.10/ctypes/__init__.py", line 374, in __init__
 self._handle = _dlopen(self._name, mode)
OSError: /home/maint/miniforge3/envs/meshcoder/lib/python3.10/site-packages/torch_scatter/_version_cuda.so: undefined symbol: _ZN5torch3jit17parseSchemaOrNameERKSsb```

[tathya-ow]

For some reason it worked when I changed my conda install pytorch==2.4.1 torchvision==0.19.1 torchaudio==2.4.1 pytorch-cuda=11.8 -c pytorch -c nvidia -y to this pip install torch==2.4.1 torchvision==0.19.1 torchaudio==2.4.1 --index-url https://download.pytorch.org/whl/cu118.

[tathya-ow]

Also, when I install pytorch3d using conda install pytorch3d -c pytorch3d, the torch and CUDA version changes 2.4.1 to 2.6.0 and 11.8 to 12.6. Is that the same thing happened to the original implementation?

However, the inference is working with these configuration and not pytorch=2.4.1 and CUDA=11.8