jiayingwu19 / decor Goto Github PK

I tried to run with -Run DECOR using RTX-4090 and got "RuntimeError: CUDA error: CUBLAS_STATUS_ALLOC_FAILED when calling cublasCreate(handle)" when using gossipcop profile.

I'm pretty sure it's an OOM situation because I'm using another profile. able to operate and
Similar results were obtained in the paper. So, is there any way to improve the above situation?

Traceback (most recent call last):
File "src/gcn_decor.py", line 130, in
acc, prec, recall, f1 = train(data, iter)
File "src/gcn_decor.py", line 100, in train
loss.backward()
File "N:\environment\Anaconda3\envs\py38_torch20_Gnn\lib\site-packages\torch_tensor.py", line 487, in backward
torch.autograd.backward(
File "N:\environment\Anaconda3\envs\py38_torch20_Gnn\lib\site-packages\torch\autograd_init_.py", line 200, in backward
Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
RuntimeError: CUDA error: CUBLAS_STATUS_ALLOC_FAILED when calling cublasCreate(handle)

Environment

• Python 3.8.10
• pytorch 2.0.0
• pyg 2.3.0
• torch-cluster 1.6.1+pt20cu118
• torch-scatter 2.1.1+pt20cu118
• torch-sparse 0.6.17+pt20cu118
• torch-spline-conv 1.2.2+pt20cu118
• torchaudio 2.0.0
• torchvision 0.15.0