Dependency Conflict: PyTorch 2.0 requires mmcv-full>=1.7.2, but mmdet3d requires mmcv-full<1.7.2

[PureKoala]

I am attempting to set up the UniAD environment using PyTorch 2.0.1 and CUDA 11.8 just as installation document. I have encountered a circular dependency issue regarding mmcv-full and mmdet3d.

Current Environment:

• Python: 3.9
• PyTorch: 2.0.1
• CUDA: 11.8

The Problem:

There is a conflict between the requirements for PyTorch 2.0 and the requirements for mmdet3d==1.0.0rc6 (which is required by UniAD).

1. Scenario A (Follow UniAD requirements):
   I try to install mmcv-full==1.6.1 (as required by mmdet3d).
   Result: Since there are no pre-compiled wheels for mmcv-full 1.6.1 on PyTorch 2.0 + CUDA 11.8, the installation fails to build CUDA extensions.
   Error: ModuleNotFoundError: No module named 'mmcv._ext'

2. Scenario B (Support PyTorch 2.0):
   I install mmcv-full==1.7.2 (compatible with PyTorch 2.0).
   Result: mmcv._ext loads correctly, but mmdet3d crashes because it strictly asserts mmcv_full < 1.7.0.
   Error: Version incompatibility assertion in mmdet3d.

Is there a recommended workaround to run UniAD with PyTorch 2.0? Or is strict downgrade to PyTorch 1.12.x the only supported path currently?