[Bug] CUDA failure: 'invalid configuration argument' when batch_size is 1 or 2
LiSu opened this issue · 1 comments
LiSu commented
🐛 Describe the bug
python train_rgnn.py
model size: 3.967MB
0%| | 0/20 [00:00<?, ?it/s]CUDA failure /codelab/graphlearn-for-pytorch/graphlearn_torch/csrc/cuda/random_sampler.cu:222: 'invalid configuration argument'
Environment
- GLT version:
- PyG version:
- PyTorch version:
- OS:
- Python version:
- CUDA/cuDNN version:
- Any other relevant information