[BUG]: colossalai check -i failed after installation both from PyPl and source

[MDLS-Delores]

🐛 Describe the bug

I encounter this bug using "colossalai check -i" right after the installation.
File "../anaconda3/envs/base/lib/python3.8/site-packages/torch/_ops.py", line 485, in getattr
raise AttributeError(
AttributeError: The underlying op of 'aten.upsample_nearest2d_backward' has no overload name 'vec'

And this can be traced to file 'anaconda3/envs/base/lib/python3.8/site-packages/colossalai/fx/profiler/opcount.py' at line 273, aten.upsample_nearest2d_backward.vec: elementwise_flop_counter(0, 1).

Environment

CUDA 12.0/PyTorch 2.0.0/Pytorch-cuda 11.8/python 3.8.16/colossalai 0.2.5

[MDLS-Delores]

It seems to be solved by changing line 273 in file 'anaconda3/envs/base/lib/python3.8/site-packages/colossalai/fx/profiler/opcount.py' from "aten.upsample_nearest2d_backward.vec: elementwise_flop_counter(0, 1)," to "aten.upsample_nearest2d_backward.default: elementwise_flop_counter(0, 1),".
It seems that in my environment, the torch.ops.aten.upsample_nearst2d_backward class does not have vec attribute.

[JThh]

torch 2.0. is not supported by CAI yet. Please try downgrading it into 1.x

[MDLS-Delores]

ok. thanks for your guidance.

[gfiameni]

Hi all,
I am experiencing the same issue with this environment: CUDA 12.0/PyTorch 1.14/Python 3.8.0/CUDA Python 12.1.0/colossalai 0.2.5

Wondering if CUDA 12.0/ColossalAI is the root cause.

Thanks!

[liyanhui607]

It seems to be solved by changing line 273 in file 'anaconda3/envs/base/lib/python3.8/site-packages/colossalai/fx/profiler/opcount.py' from "aten.upsample_nearest2d_backward.vec: elementwise_flop_counter(0, 1)," to "aten.upsample_nearest2d_backward.default: elementwise_flop_counter(0, 1),". It seems that in my environment, the torch.ops.aten.upsample_nearst2d_backward class does not have vec attribute.

this suggestion can work

[Camille7777]

Hi all, I am experiencing the same issue with this environment: CUDA 12.0/PyTorch 1.14/Python 3.8.0/CUDA Python 12.1.0/colossalai 0.2.5

Wondering if CUDA 12.0/ColossalAI is the root cause.

Thanks!

torch==1.13.1 can work

[XuJing1022]

When can ColossalAI support Pytorch2.0? T T

[FrankLeeeee]

When can ColossalAI support Pytorch2.0? T T

Plan to start working on it in April :) Do you have any urgent feature request?