Uni-Mol
Uni-Mol copied to clipboard
fused_multi_tensor is not installed corrected
When "from unimol_tools import UniMolRepr", I get the following messages:
fused_multi_tensor is not installed corrected fused_rounding is not installed corrected fused_layer_norm is not installed corrected fused_softmax is not installed corrected
I would like to know how to solve this issue or if it has any impact on using UniMol normally?
It seems missing uni-core, you can follow this.
Hello, I'm having the same problem but I have already installed Uni-Core following the instructions from their GitHub. `>>> from unimol_tools import UniMolRepr fused_multi_tensor is not installed corrected fused_rounding is not installed corrected fused_layer_norm is not installed corrected fused_softmax is not installed corrected
clf = UniMolRepr(data_type='molecule') 2023-12-05 22:46:32 | unimol_tools/models/unimol.py | 116 | INFO | Uni-Mol(QSAR) | Loading pretrained weights from /home/yutonglu/.local/lib/python3.10/site-packages/unimol_tools-1.0.0-py3.10.egg/unimol_tools/weights/mol_pre_all_h_220816.pt smiles = ['CCO', 'CCC', 'CCCC'] reprs = clf.get_repr(smiles) 2023-12-05 22:52:12 | unimol_tools/data/conformer.py | 62 | INFO | Uni-Mol(QSAR) | Start generating conformers... 3it [00:01, 2.54it/s] 2023-12-05 22:52:13 | unimol_tools/data/conformer.py | 66 | INFO | Uni-Mol(QSAR) | Failed to generate conformers for 0.00% of molecules. 2023-12-05 22:52:13 | unimol_tools/data/conformer.py | 68 | INFO | Uni-Mol(QSAR) | Failed to generate 3d conformers for 0.00% of molecules. 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 1.29it/s] `
I was not able to reproduce the example. I was wondering if this is related to the missing fused_multi_tensor, etc. Thanks!
Hi, fused_multi_tensor
is cuda kernel in unicore, the warning is said that your version didn't compiled with the correct cuda / pytorch version, If you want to use the cuda kernel in your custom envrioment, you can compile it from source python setup.py install
. More details about installation can be found here.
Hi! I have installed Unicore according to the instructions in https://nb.bohrium.dp.tech/detail/3074300073. I make sure the wheel has the correct Cuda version, PyTorch version, and Python version. But the problem persists. I still got the warning message: fused_multi_tensor is not installed corrected fused_rounding is not installed corrected fused_layer_norm is not installed corrected fused_softmax is not installed corrected
@lindafei01 what compile option do you used ? If you install it from source, the warning message still raise ?
Hi! My installation process is shown in the attached png.
unimol-tool.pdf It seems that I fail to upload a PNG file. The installation process is attached in this pdf. Thank you!
I am also trying to install Unicore from the source. I would like to know how to specify the version. Running 'python setup.py install' by default requires torch>=2.0.0.
I am also trying to install Unicore from the source. I would like to know how to specify the version. Running 'python setup.py install' by default requires torch>=2.0.0.
Yes, now uni-core requires torch >= 2.0 by default. If you insist installing previous version, you need checkout the commit before that. For better experience, we recommend you to upgrade the pytorch to 2.0 or above.
Hi, I also encountered the same problem:fused_multi_tensor is not installed corrected fused_rounding is not installed corrected fused_layer_norm is not installed corrected fused_softmax is not installed corrected And Finally, it also warned:ValueError: Cannot register duplicate model (unimol) Ask for help, thanks!