AttributeError: module 'torch.compiler' has no attribute 'is_compiling'

[LukeLIN-web]

Describe the Bug

Minimal Steps/Code to Reproduce the Bug

.conda/envs/opensora/lib/python3.9/site-packages/apex/normalization/fused_layer_norm.py", line 412, in forward if torch.jit.is_tracing() or torch.jit.is_scripting() or torch.compiler.is_compiling() or not input.is_cuda: AttributeError: module 'torch.compiler' has no attribute 'is_compiling'

pip install -v --disable-pip-version-check --no-cache-dir --no-build-isolation --config-settings "--build-option=--cpp_ext" --config-settings "--build-option=--cuda_ext" git+https://github.com/NVIDIA/apex.git I installed torch 2.2.2

Environment

apex 0.1
Versions of relevant libraries: [pip3] galore-torch==1.0 [pip3] numpy==1.26.4 [pip3] rotary-embedding-torch==0.5.3 [pip3] torch==2.2.2 [pip3] torchvision==0.17.2 [pip3] triton==2.2.0 [conda] galore-torch 1.0 pypi_0 pypi [conda] numpy 1.26.4 pypi_0 pypi [conda] rotary-embedding-torch 0.5.3 pypi_0 pypi [conda] torch 2.2.2 pypi_0 pypi [conda] torchvision 0.17.2 pypi_0 pypi [conda] triton 2.2.0 pypi_0 pypi

[venom8898]

I am also running in to this. I also had to had to upgrade diffusers to 0.31.0 to even get the models to download. But now when I try their multigpu example I run in to the "AttributeError: module 'torch.compiler' has no attribute 'is_compiling'" error

[kkahatapitiya]

Instead of using the master, checking-out a previous tag (https://github.com/NVIDIA/apex/tree/24.04.01) before building, solved this for me.

[zyang6]

Instead of using the master, checking-out a previous tag (https://github.com/NVIDIA/apex/tree/24.04.01) before building, solved this for me. Have you solved this? Are you compiling apex packages before 24.01.01?

[kkahatapitiya]

Yes, using the version with tag 24.01.01 (git checkout tags/24.01.01), and building from source solved this issue for me.

[zyang6]

Yes, using the version with tag 24.01.01 (git checkout tags/24.01.01), and building from source solved this issue for me.

Thank you !I'll give it a try

[yinghuozijin]

Yes, using the version with tag 24.01.01 (git checkout tags/24.01.01), and building from source solved this issue for me.

Correcting the typo: 24.04.01