[Colab] Torch necessary versions
m1litaru opened this issue · comments
Hello again, @zachteed, @jiadeng @heilaw @troylhy1991!
Can you please specify what versions are required for torch, torchvision and torchaudio? I have been trying different combinations of versions these days (like the command below), but none of them worked.
!pip install torch==1.10.1+cu111 torchvision==0.11.2+cu111 torchaudio==0.10.1+cu111 -f https://download.pytorch.org/whl/torch_stable.html
Please help me as soon as you have time!
Below you can find the error I can't get rid of.
File "scripts/demo.py", line 78, in demo Ts = model(image1, image2, depth1, depth2, intrinsics, iters=16) File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 1102, in _call_impl return forward_call(*input, **kwargs) File "/usr/local/lib/python3.7/dist-packages/torch/nn/parallel/data_parallel.py", line 166, in forward return self.module(*inputs[0], **kwargs[0]) File "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py", line 1102, in _call_impl return forward_call(*input, **kwargs) File "./raft3d/raft3d.py", line 161, in forward coords1_xyz, _ = pops.projective_transform(Ts, depth1_r8, intrinsics_r8) File "./raft3d/projective_ops.py", line 43, in projective_transform x1 = project(X1, intrinsics) File "./raft3d/projective_ops.py", line 13, in project x = fx * (X / Z) + cx **RuntimeError: CUDA error: no kernel image is available for execution on the device**
The requirements from lietorch repo mention that the code should work with any versions satisfying the following conditions:
- Cuda >= 10.1 (with nvcc compiler)
- PyTorch >= 1.8.
which is indeed false, the code working only (in my environment, on Google Colaboratory) with those exact versions: Cuda==10.01 and Pythorch ==1.8.
(Please make sure you have uninstalled other versions of Cuda before executing the following instructions!)
!sudo apt update
!sudo apt install cuda-10-1
!sudo apt install libcudnn7
!pip install torch==1.8.0