PyPI Versions and Compatibility with Torch

[ziqihuangg]

No description provided.

[ziqihuangg]

We address some common questions recently shared by our users (thanks for the feedback!). A fix will be included in the upcoming vbench 0.1.5 release.

Q1: Why does pip install vbench install version 0.1.0 instead of 0.1.4?
A1: That's probably because torch 2.0.0 (or above) is already installed in your environment. Only vbench 0.1.0 allows torch 2.0.0 and above. vbench 0.1.1 to vbench 0.1.4 requires torch<2.0.0 and >=1.12. This strict dependency on Torch version is to ensure reproducibility of numerical results in some scenarios. We will try to relax this restriction to improve compatibility with newer Torch versions, and provide the fix in vbench 0.1.5.

Q2: When using vbench installed via pip, why do I encounter errors like unrecognized arguments: --mode=custom_input?
A2: That's because the vbench package is not up-to-date (see Q1 above for why).